Bedrock

Amazon Bedrock Agents & Knowledge Bases – Complete Guide [2026]

~ Last updated on : ~ Kiro Agent

Amazon Bedrock Agents, Knowledge Bases & Guardrails – Complete Guide

Amazon Bedrock provides a comprehensive platform for building, deploying, and managing generative AI applications. This deep-dive guide covers the advanced capabilities of Bedrock’s key components: Knowledge Bases for RAG, Agents for autonomous task execution, AgentCore for production deployment, Guardrails for safety, Model Evaluation, Fine-tuning, and Prompt Management.

Amazon Bedrock Knowledge Bases

Amazon Bedrock Knowledge Bases is a fully managed RAG (Retrieval-Augmented Generation) capability that connects foundation models to proprietary data sources. It handles the entire workflow from data ingestion, chunking, embedding, storage, to retrieval and prompt augmentation.

Knowledge Base Types

  • Custom Knowledge Base – You choose the vector store, embedding model, chunking strategy, and data sources. Provides full control over the RAG pipeline.
  • Managed Knowledge Base (GA June 2026) – Amazon Bedrock manages the underlying infrastructure including vector storage, embeddings, re-ranking, and retrieval optimization. Supports auto-scaling, agentic retrieval for multi-hop reasoning, and multimodal data ingestion.

Data Sources

  • Amazon S3 – Primary data source supporting documents in PDF, TXT, MD, HTML, CSV, DOC/DOCX, XLS/XLSX, and JSON formats.
  • Confluence – Connects to Atlassian Confluence workspaces for ingesting wiki pages and documentation.
  • Microsoft SharePoint – Ingests documents from SharePoint Online sites and libraries.
  • Salesforce – Connects to Salesforce objects like Knowledge Articles and custom objects.
  • Web Crawler – Crawls and ingests web pages from specified URLs with configurable depth and scope.
  • Google Drive – Connects to Google Drive for document ingestion (Managed KB).
  • OneDrive – Connects to Microsoft OneDrive (Managed KB).

Chunking Strategies

  • Default Chunking – Splits content into chunks of approximately 300 tokens, honoring sentence boundaries.
  • Fixed-size Chunking – Splits content into chunks of a user-defined token size (1–8192 tokens) with configurable overlap percentage for context continuity.
  • Semantic Chunking – Groups text by meaning using embedding similarity. Breakpoints are created when semantic similarity between consecutive sentences drops below a threshold. Produces more coherent chunks but is computationally more expensive.
  • Hierarchical Chunking – Creates parent-child chunk relationships. Parent chunks provide broader context while child chunks contain specific details. During retrieval, child chunks are returned with parent context for better comprehension.
  • No Chunking – Treats each document as a single chunk. Best for short documents or pre-chunked data.

Embedding Models

  • Amazon Titan Text Embeddings V2 – AWS native model supporting configurable output dimensions (256, 512, or 1024). Supports text normalization and multiple languages. Optimized for RAG workloads with high accuracy-to-cost ratio.
  • Cohere Embed – Multilingual embedding model available in English and multilingual variants. Supports input types (search_document, search_query) for optimized retrieval.
  • Amazon Titan Multimodal Embeddings – Supports both text and image embeddings in a unified vector space.

Vector Stores

  • Amazon OpenSearch Serverless – Default option with serverless scaling. Supports hybrid search (semantic + keyword), metadata filtering, and automatic index management.
  • Amazon OpenSearch Service (Managed Cluster) – Added March 2025. Provides more control over cluster configuration, instance types, and scaling policies.
  • Amazon Aurora PostgreSQL – Uses pgvector extension. Supports hybrid search (added April 2025) and integrates with existing Aurora databases.
  • Pinecone – Third-party managed vector database with serverless and pod-based options.
  • Redis Enterprise Cloud – In-memory vector store for low-latency retrieval.
  • MongoDB Atlas – Document database with vector search capabilities. Supports hybrid search (added April 2025).
  • Amazon Neptune Analytics – Graph + vector search for knowledge graph use cases.
  • Amazon S3 – Added July 2025 for cost-effective vector storage with S3-native retrieval.

Hybrid Search

  • Combines semantic (vector) search with keyword (lexical) search for improved retrieval accuracy.
  • Semantic search captures meaning and handles paraphrasing; keyword search handles exact matches, names, and codes.
  • Supported on OpenSearch Serverless, OpenSearch Managed Clusters, Aurora PostgreSQL (April 2025), and MongoDB Atlas (April 2025).
  • Results are combined using Reciprocal Rank Fusion (RRF) to produce a unified ranking.

Metadata Filtering

  • Each document can have a metadata JSON file (up to 10 KB) with custom attributes.
  • Filters are applied as pre-filtering before vector search, reducing the search space.
  • Supports operators: equals, notEquals, greaterThan, lessThan, in, notIn, startsWith, stringContains.
  • Enables multi-tenant RAG by filtering documents based on tenant ID, access controls, or document categories.

Advanced Parsing

  • Foundation Model Parsing – Uses an FM (e.g., Claude) to extract and interpret content from complex documents including PDFs with tables, charts, and images. Provides customizable extraction prompts.
  • Amazon Textract Parsing – OCR-based parsing for scanned documents and images.
  • Standard Parsing – Default text extraction for supported document formats.
  • FM parsing is ideal for documents with complex layouts, embedded images, or non-standard formatting that standard parsers cannot handle accurately.

📖 Deep Dive Guides: Bedrock vs SageMaker | RAG Architecture | Prompt Engineering | Responsible AI | AI Services Decision Guide

Amazon Bedrock Agents

Amazon Bedrock Agents enables developers to build autonomous AI agents that can plan multi-step tasks, invoke APIs, and interact with knowledge bases to accomplish complex goals. Agents use foundation models for reasoning and orchestration.

Agent Architecture

  • Foundation Model – The reasoning engine that interprets user requests, plans actions, and generates responses.
  • Instructions – System-level prompts that define the agent’s persona, capabilities, and behavioral guidelines.
  • Action Groups – Collections of tools/APIs the agent can invoke, defined via OpenAPI schemas or function definitions.
  • Knowledge Bases – Connected data sources for RAG-based retrieval to ground responses in proprietary data.
  • Guardrails – Safety filters applied to agent inputs and outputs.

Orchestration

  • Agents use a ReAct (Reasoning + Acting) orchestration loop by default: the FM reasons about the task, decides on an action, executes it, observes results, and iterates.
  • Custom Orchestration – Use a Lambda function to define custom orchestration logic, overriding the default ReAct loop for specialized workflows.
  • The orchestration loop continues until the agent determines it has sufficient information to generate a final response or reaches the maximum iteration limit.

Action Groups & Tool Use

  • Action groups define the tools available to the agent using either OpenAPI schemas or simplified function definitions.
  • Lambda Functions – Backend logic executed when the agent invokes an action. Receives the API operation, parameters, and session context.
  • Return of Control (ROC) – Instead of executing a Lambda, the agent returns control to the calling application with the action details. The application executes the action and returns results to continue the conversation.
  • Code Interpreter – Built-in action group that allows the agent to generate and execute Python code in a secure sandbox for data analysis, calculations, and chart generation.
  • User Confirmation – Configurable step where the agent asks for user approval before executing sensitive actions.

Multi-Step Reasoning

  • Agents decompose complex requests into sequential sub-tasks, executing each step and using results to inform the next.
  • Supports query decomposition for knowledge base retrieval – breaking a complex question into simpler sub-queries.
  • Chain-of-thought traces are available for debugging and observability.

Inline Agents

  • Dynamically configure agent capabilities at runtime without pre-creating agent resources.
  • Specify instructions, action groups, knowledge bases, and guardrails in the API call itself.
  • Enables dynamic workflow adaptation where agent roles and tools change based on context.
  • Launched with multi-agent collaboration GA (March 2025).

Multi-Agent Collaboration (Supervisor/Child)

  • Supervisor Agent – Orchestrates the workflow by breaking requests into sub-tasks and delegating to specialized child agents.
  • Child Agents (Collaborator Agents) – Specialized agents focused on specific domains (e.g., checking maintenance, analyzing alarms, evaluating KPIs).
  • Supervisor routes tasks, consolidates outputs, and generates unified final responses.
  • Supports both SUPERVISOR mode (supervisor decides routing) and SUPERVISOR_ROUTER mode (classifier-based routing).
  • GA since March 2025 with support for up to 5 collaborator agents per supervisor.

Agent Memory

  • Session Memory (Short-term) – Maintains conversation context within a session. Automatically managed within the session window (configurable idle timeout).
  • Long-term Memory – Persists information across sessions. Extracts key facts, preferences, and context from conversations and stores them for future sessions.
  • Memory enables personalized experiences where agents remember user preferences, past interactions, and ongoing tasks.
  • Supports metadata on memory records for organizing, filtering, and routing retrieval.

Prompt Engineering for Agents

  • System Instructions – Define the agent’s role, personality, constraints, and response format.
  • Advanced Prompts – Customize prompts at each orchestration step: pre-processing, orchestration, knowledge base response generation, and post-processing.
  • Prompt Templates – Use variables (e.g., $tool_results$, $knowledge_base_results$) to structure how the agent processes information.
  • Best practices: Be specific about capabilities, define clear boundaries, provide examples of expected behavior, and specify output formats.

Amazon Bedrock AgentCore

Amazon Bedrock AgentCore (GA June 2026) is a code-first platform to build, deploy, connect, and optimize AI agents at scale. It provides production-grade infrastructure including runtime, identity, tools, memory, observability, and evaluation — regardless of the framework or model used.

Managed Deployment (AgentCore Runtime & Harness)

  • AgentCore Harness – The managed orchestration layer (“body”) for agents. Handles the orchestration loop, tool execution, context window management, state persistence, failure recovery, and session isolation.
  • Define agents via configuration: model, tools, skills, instructions. AgentCore assembles and runs the agent loop.
  • Each agent runs in its own isolated environment with filesystem, shell, memory, and web browsing capabilities.
  • Supports any open-source framework (LangGraph, CrewAI, Strands) and any model.
  • Provides MicroVM-based isolation for secure execution of tools and code.

AgentCore Identity & Access

  • AgentCore Identity – Provides robust identity and access management for agents at scale.
  • Agents can access resources/tools on behalf of users or themselves with pre-authorized user consent.
  • Compatible with existing identity providers (Okta, Auth0, Entra ID) — no user migration required.
  • Workload Identities – Unique identities assigned to agents for authentication and authorization.
  • Centralized identity management regardless of deployment environment (AgentCore Runtime, self-hosted, hybrid).
  • Eliminates need for custom access controls and identity infrastructure.

Tool Management (AgentCore Gateway)

  • AgentCore Gateway – Unified MCP (Model Context Protocol) gateway for tool discovery and invocation.
  • Serves as a single endpoint for accessing tools from different teams, organizations, and applications.
  • Fine-grained access control with gateway interceptors for per-principal permissions.
  • Supports the AWS-curated skills catalog accessible with a single toggle.
  • Web Search tool enables agents to ground responses in current web knowledge.

Memory Management

  • Memory provisions automatically when a harness is created.
  • Extracts useful information from short-term memory and stores as long-term memory records.
  • Supports strictly consistent metadata on memory records for organized retrieval.
  • Agents recognize returning users without additional setup.

Quality Evaluations

  • Batch Evaluation – Define what “good” looks like and measure candidate changes against quality bars at scale.
  • Customers specify evaluation criteria and AgentCore runs assessments across multiple test cases.
  • Supports comparison of agent versions before deployment.

A/B Testing

  • Controlled comparison between agent versions by splitting live production traffic.
  • Measures outcomes side-by-side to confirm improvements hold under real conditions.
  • Enables data-driven decisions about agent updates and configuration changes.

Policy Controls

  • AgentCore Policy – Authorization capability that controls which actions agents are authorized to take.
  • Integrates with Amazon Bedrock Guardrails for content safety and prompt injection protection.
  • Provides enterprise defenses against security and safety risks in agent workloads.
  • Supports sensitive data exposure prevention and prompt injection attack detection.

Amazon Bedrock Guardrails

Amazon Bedrock Guardrails provides configurable safeguards for generative AI applications. It helps detect and filter harmful content, block undesirable topics, redact sensitive information, and reduce hallucinations — applied to both user inputs and model responses.

Content Filters

  • Detect and filter harmful content across six categories with configurable strength levels (None, Low, Medium, High):
  • Hate – Content that discriminates, criticizes, insults, or dehumanizes based on identity attributes.
  • Insults – Content that demeans, bullies, or includes negative/derogatory language.
  • Sexual – Content that indicates sexual interest, activity, or arousal.
  • Violence – Content that glorifies or threatens physical harm to individuals or groups.
  • Misconduct – Content related to criminal activity, including fraud, theft, and illegal substance use.
  • Prompt Attacks – Detects prompt injection and jailbreak attempts designed to bypass safety controls.
  • Supports tiered filtering (announced June 2025) for cost-optimized content moderation at scale.

Image Content Filters (GA March 2025)

  • Extends content filtering to image modality — moderates both image and text content.
  • Applies to all categories: hate, insults, sexual, violence, misconduct, and prompt attacks.
  • Blocks up to 88% of harmful multimodal content.
  • Industry-leading safeguards for applications handling user-uploaded images or model-generated images.

Denied Topics

  • Define custom topics that the AI should refuse to engage with.
  • Provide a natural language definition and optional sample phrases for each denied topic.
  • Example: A bank’s AI assistant can deny conversations about investment advice or cryptocurrencies.
  • Applied to both user inputs (block the question) and model outputs (block the response).

Word Filters

  • Block specific words or phrases from appearing in inputs or outputs.
  • Supports exact match and managed word lists (e.g., profanity lists).
  • Useful for blocking competitor names, internal project codes, or inappropriate terminology.

Sensitive Information Filters

  • PII Detection – Identifies personally identifiable information including names, email addresses, phone numbers, SSNs, credit card numbers, and more.
  • Regex Patterns – Define custom patterns for domain-specific sensitive data (e.g., account numbers, internal IDs).
  • Actions: Block (reject the entire message) or Anonymize/Redact (mask the PII and allow the message through).
  • Supports over 30 built-in PII entity types.

Contextual Grounding Check

  • Detects hallucinations in RAG and summarization use cases.
  • Grounding – Validates that model responses are factually consistent with the provided reference source/context.
  • Relevance – Checks that the response is relevant to the user’s query.
  • Configurable thresholds for grounding and relevance scores.
  • Filters over 75% of hallucinated responses in RAG applications.

Automated Reasoning Checks

  • Uses formal verification methods grounded in mathematical logic to validate AI-generated outputs.
  • Detects hallucinations, suggests corrections, and highlights unstated assumptions.
  • Provides provably correct, auditable assessments with deterministic formal logic.
  • First and only safeguard using Automated Reasoning to prevent factual errors.
  • Policy refinement workflows added June 2026 for iterative improvement.

ApplyGuardrail API

  • Standalone API to apply guardrails independently of model invocation.
  • Enables guardrail evaluation on any text content — even from non-Bedrock models or external systems.
  • Use cases: validate content from third-party LLMs, pre-screen user inputs, post-process outputs from any source.
  • InvokeGuardrailChecks API – Enhanced API for agentic AI applications requiring step-level guardrail checks.

Code Domain Support (Jan 2025)

  • Protects against undesirable content within code elements.
  • Inspects user prompts, comments, variables, function names, and string literals.
  • Prevents injection of harmful content via code constructs.

Amazon Bedrock Model Evaluation

Amazon Bedrock Evaluations helps you compare, evaluate, and select foundation models for your specific use cases. It supports automatic evaluation, human evaluation, and LLM-as-a-judge workflows.

Automatic Evaluation

  • Evaluate models using built-in metrics without human involvement.
  • Accuracy – Measures correctness of model responses using metrics like BERTScore, ROUGE, and exact match.
  • Robustness – Tests model consistency across paraphrased inputs and adversarial perturbations.
  • Toxicity – Measures harmful or inappropriate content in model outputs.
  • Supports custom datasets in JSONL format with prompt-response-reference triples.
  • Can evaluate models running on Bedrock, other cloud providers, or on-premises (GA April 2025).

LLM-as-a-Judge (Preview Dec 2024)

  • Uses a foundation model to evaluate other models with human-like quality assessment.
  • Fraction of the cost and time of human evaluations.
  • Supports custom evaluation criteria and scoring rubrics.

RAG Evaluation

  • Evaluate end-to-end RAG systems including retrieval quality and generation accuracy.
  • Metrics: context relevance, answer faithfulness, answer relevance.
  • Can evaluate fully built applications, not just individual model responses.

Human Evaluation Workflows

  • Set up human evaluation jobs with custom work teams.
  • Evaluators rate model responses on custom criteria (helpfulness, harmlessness, coherence).
  • Supports comparison of multiple models side-by-side.
  • Integrates with Amazon SageMaker Ground Truth for workforce management.

Model Comparison

  • Compare multiple foundation models on the same evaluation dataset.
  • Side-by-side results with statistical significance testing.
  • Helps select optimal model balancing quality, latency, and cost for specific use cases.

Amazon Bedrock Fine-Tuning & Customization

Amazon Bedrock provides multiple model customization techniques to adapt foundation models to specific tasks and domains.

Continued Pre-Training

  • Extend a model’s knowledge by training on unlabeled, domain-specific data.
  • Adapts the model’s language understanding to specialized vocabularies and concepts.
  • Training data format: Plain text documents in S3 (no prompt-completion pairs needed).
  • Best for: Domain adaptation (medical, legal, financial terminology).

Instruction Fine-Tuning

  • Train models on labeled prompt-completion pairs to improve task-specific performance.
  • Training data format: JSONL with {"prompt": "...", "completion": "..."} or chat-format messages.
  • Supports validation datasets for monitoring overfitting.
  • Configurable hyperparameters: epochs, batch size, learning rate, warmup steps.
  • Best for: Improving performance on specific tasks like classification, extraction, or formatting.

Reinforcement Fine-Tuning (RFT) – GA December 2025

  • Advanced customization using reward-based learning without requiring large labeled datasets.
  • Bring your own prompts or use existing Bedrock API invocation logs as training data.
  • Delivers 66% accuracy gains on average over base models.
  • Supported models: Amazon Nova, OpenAI GPT OSS 20B, Qwen 3 32B (Feb 2026).
  • Automates the reinforcement workflow — accessible to developers without deep ML expertise.
  • Built-in evaluation tools to compare RFT model against the base model.
  • Supports iterative fine-tuning: build upon previously customized models for continuous improvement.
  • Training data: JSONL with prompts; rewards are computed by a verifier/judge function you define.

Model Distillation

  • Transfer knowledge from a larger “teacher” model to a smaller “student” model.
  • Provide input prompts in JSONL; Bedrock generates responses from the teacher model and uses them to fine-tune the student.
  • Achieves teacher-model quality at student-model cost and latency.
  • Best for: Reducing inference costs while maintaining quality for specific use cases.

Training Data Format Summary

Method Data Format Data Requirements
Continued Pre-Training Plain text files Unlabeled domain corpus
Instruction Fine-Tuning JSONL (prompt/completion) Min ~100 examples, recommended 1000+
Reinforcement Fine-Tuning JSONL (prompts) + verifier Prompts + reward/judge function
Distillation JSONL (input prompts) Prompts only; teacher generates completions

Amazon Bedrock Prompt Flows & Management

Amazon Bedrock provides tools for creating, managing, and orchestrating prompts and generative AI workflows.

Prompt Management (GA November 2024)

  • Streamlined interface to create, evaluate, version, and share prompts.
  • Prompt Versioning – Each version is linked to its evaluation results. Supports rollback, audit trails, and A/B testing.
  • Prompt Variables – Template variables (e.g., {{context}}, {{question}}) for dynamic prompt construction.
  • Model Selection – Test the same prompt across different foundation models to compare performance.
  • Sharing – Share prompts across teams and projects for collaboration and reuse.
  • Treats prompts as critical as code — version-controlled and reproducible.

Bedrock Flows (Visual Flow Builder)

  • Intuitive visual builder to create, test, and deploy generative AI workflows.
  • Drag-and-drop interface to link Prompts, Agents, Knowledge Bases, Guardrails, and AWS services.
  • Node Types:
    • Prompt Node – Invokes a foundation model with a configured prompt.
    • Agent Node – Invokes a Bedrock Agent for autonomous task execution.
    • Knowledge Base Node – Retrieves relevant information from a Knowledge Base.
    • Condition Node – Routes flow based on conditional logic.
    • Lambda Node – Executes custom business logic.
    • Lex Node – Integrates with Amazon Lex for conversational interfaces.
    • Iterator Node – Loops over collections of items.
    • Collector Node – Aggregates results from parallel or iterated executions.
  • Serverless execution — pricing based on resources consumed (model invocations, Lambda, etc.).
  • Supports versioning and aliases for deployment management.

A/B Testing for Prompts

  • Version-control prompts and compare performance across versions.
  • Use Bedrock Evaluations to measure quality differences between prompt versions.
  • Deploy prompt versions with aliases and switch traffic between versions.
  • Combine with AgentCore A/B testing for full agent-level experimentation.

Comparison: Bedrock Knowledge Bases vs Amazon Kendra vs OpenSearch

Feature Bedrock Knowledge Bases Amazon Kendra Amazon OpenSearch Service
Primary Purpose RAG for generative AI Intelligent enterprise search Full-text search, analytics, vector search
Search Type Semantic + hybrid (keyword) Semantic + keyword (NLU-based) Full-text, keyword, vector (k-NN), hybrid
RAG Integration Native (fully managed) Via Retrieve API + custom orchestration Custom implementation required
Management Fully managed Fully managed Managed clusters or serverless
Data Sources S3, Confluence, SharePoint, Salesforce, Web Crawler, Google Drive, OneDrive 40+ connectors (S3, SharePoint, Salesforce, databases, ServiceNow, etc.) Custom ingestion pipelines
Chunking Fixed, semantic, hierarchical, no chunking Automatic (document passages) Custom (application-managed)
Vector Store Managed or BYO (OpenSearch, Aurora, Pinecone, Redis, MongoDB) Built-in (not configurable) Native k-NN plugin
Metadata Filtering Yes (custom JSON metadata) Yes (document attributes) Yes (field-level filtering)
Access Control Via metadata filtering Native ACL integration (SharePoint, etc.) Fine-grained access control
Multimodal Yes (FM parsing for images/tables) Limited (document text extraction) Yes (with custom embeddings)
Re-ranking Yes (Managed KB) Built-in semantic re-ranking Custom (Learning to Rank plugin)
Best For GenAI applications, RAG pipelines, AI agents Enterprise search portals, FAQ systems, document discovery Custom search, log analytics, observability, full control over retrieval
Pricing Model Pay per query + storage (vector store) Index-based (provisioned capacity) Instance/serverless OCU hours

AWS Certification Exam Practice Questions

Question 1:

A company is building a RAG application using Amazon Bedrock Knowledge Bases. Their documents contain complex tables, charts, and embedded images in PDF format. Standard text extraction is losing critical information. Which parsing approach should they use to improve data quality?

  1. Fixed-size chunking with 512 tokens
  2. Foundation model parsing with a customized extraction prompt
  3. Semantic chunking with sentence boundary detection
  4. Amazon Textract with default settings
Show Answer

Answer: B – Foundation model parsing uses an FM (e.g., Claude) to interpret complex document layouts including tables, charts, and images. It allows customizable extraction prompts to capture the specific information needed. While Textract handles OCR, FM parsing provides superior understanding of document structure and semantics.

Question 2:

A financial services company wants their Bedrock Agent to execute a trade only after receiving explicit user approval. Which feature should they implement?

  1. Guardrails with denied topics
  2. Return of Control with user confirmation
  3. Custom orchestration with Lambda
  4. Multi-agent collaboration with a supervisor
Show Answer

Answer: B – Return of Control (ROC) allows the agent to return the proposed action to the calling application instead of executing it directly. Combined with user confirmation configuration, this ensures sensitive actions like trade execution require explicit user approval before proceeding.

Question 3:

An organization is deploying multiple AI agents that need to access different enterprise tools and data sources on behalf of users. Each agent requires its own identity with scoped permissions and integration with their existing Okta identity provider. Which service should they use?

  1. Amazon Bedrock Agents with IAM roles
  2. Amazon Bedrock AgentCore Identity
  3. AWS IAM Identity Center with SAML federation
  4. Amazon Cognito User Pools
Show Answer

Answer: B – Amazon Bedrock AgentCore Identity provides robust identity and access management for agents at scale. It’s compatible with existing identity providers (including Okta) without requiring user migration, assigns unique workload identities to agents, and provides centralized identity management regardless of deployment environment.

Question 4:

A healthcare company uses Amazon Bedrock to generate patient-facing content. They need to ensure responses don’t contain hallucinated medical information and are always grounded in the reference documents provided. Which Guardrails feature provides the MOST reliable hallucination detection?

  1. Content filters set to High
  2. Contextual grounding check
  3. Denied topics for medical advice
  4. Automated Reasoning checks
Show Answer

Answer: D – Automated Reasoning checks use formal verification methods grounded in mathematical logic to validate AI-generated outputs. They provide provably correct, auditable assessments and can detect hallucinations, suggest corrections, and highlight unstated assumptions — making them the most reliable option for critical healthcare content. Contextual grounding is useful but probabilistic, while Automated Reasoning is deterministic.

Question 5:

A company wants to improve their foundation model’s performance on a specific classification task but has limited labeled data (only 50 examples). They do have access to a high-quality larger model and 5,000 unlabeled prompts representative of their use case. Which customization approach is MOST appropriate?

  1. Instruction fine-tuning with the 50 labeled examples
  2. Continued pre-training with domain documents
  3. Model distillation using the larger model as teacher
  4. Reinforcement fine-tuning with a reward function
Show Answer

Answer: C – Model distillation transfers knowledge from a larger “teacher” model to a smaller “student” model. The company provides their 5,000 unlabeled prompts, Bedrock generates high-quality responses from the teacher model, and uses those to fine-tune the student. This achieves teacher-model quality at lower cost without requiring labeled data. With only 50 labeled examples, instruction fine-tuning would likely underperform.

Frequently Asked Questions

What is a Bedrock Knowledge Base?

A Bedrock Knowledge Base connects your data sources (S3, web pages, Confluence, etc.) to foundation models via RAG. It automatically chunks documents, generates embeddings, stores them in a vector database, and retrieves relevant context to ground model responses in your data.

What are Bedrock Guardrails?

Guardrails are configurable safety controls that filter harmful content, block denied topics, mask PII, and verify response grounding. They can be applied to any Bedrock model call, agent, or knowledge base to ensure responsible AI usage within your organization’s policies.

How do Bedrock Agents work?

Bedrock Agents use a foundation model to break down user requests into steps, determine which tools/APIs to call (action groups), execute them, and synthesize results. They support multi-step reasoning, code execution, memory across sessions, and can collaborate with other agents.

References

AWS AI & ML Services Cheat Sheet – AIF-C01 & AIP-C01

~ Last updated on : ~ Kiro Agent

AWS AI & Generative AI Services – Cheat Sheet

This is the definitive cheat sheet covering AI, Machine Learning, and Generative AI services on AWS — designed as the anchor page for both the AWS Certified AI Practitioner (AIF-C01) and AWS Certified Generative AI Developer – Professional (AIP-C01) exams.

Related Posts:

AI/ML/Generative AI Fundamentals

AI vs ML vs Deep Learning vs Generative AI

Concept Definition Examples
Artificial Intelligence (AI) Broad field of computer science focused on creating systems that can perform tasks requiring human intelligence Rule-based systems, expert systems, robotics
Machine Learning (ML) Subset of AI where systems learn from data without being explicitly programmed Fraud detection, recommendations, forecasting
Deep Learning (DL) Subset of ML using neural networks with multiple layers (deep neural networks) to learn complex patterns Image recognition, NLP, speech recognition
Generative AI (GenAI) Subset of DL that creates new content (text, images, code, video, audio) by learning patterns from training data ChatGPT, DALL-E, Amazon Nova, Claude

Learning Paradigms

  • Supervised Learning — model learns from labeled data (input-output pairs). Used for classification (spam/not spam) and regression (price prediction).
  • Unsupervised Learning — model finds patterns in unlabeled data. Used for clustering (customer segmentation), anomaly detection, and dimensionality reduction.
  • Semi-supervised Learning — combines small amount of labeled data with large amounts of unlabeled data.
  • Reinforcement Learning (RL) — agent learns by interacting with an environment, receiving rewards/penalties. Used for game playing, robotics, and RLHF in LLMs.
  • Self-supervised Learning — model generates its own labels from input data (e.g., predicting masked tokens). Used for pre-training foundation models.

Neural Networks Basics

  • Neurons/Nodes — basic computation units that receive inputs, apply weights, add bias, and pass through an activation function.
  • Layers — Input layer (receives data), Hidden layers (process data), Output layer (produces result).
  • Weights & Biases — parameters learned during training that determine the model’s behavior.
  • Activation Functions — introduce non-linearity (ReLU, Sigmoid, Softmax, Tanh).
  • Backpropagation — algorithm to compute gradients and update weights by propagating errors backward.
  • Loss Function — measures how far the model’s predictions are from actual values.
  • Transformer Architecture — foundation of modern LLMs; uses self-attention mechanism to process entire sequences in parallel (introduced in “Attention is All You Need” paper, 2017).
  • CNNs (Convolutional Neural Networks) — specialized for image/spatial data.
  • RNNs/LSTMs — sequential data processing (largely superseded by Transformers for NLP).
  • GANs (Generative Adversarial Networks) — generator + discriminator for image generation.
  • Diffusion Models — generate images/video by learning to denoise (e.g., Stable Diffusion, Nova Canvas).

📖 Deep Dive Guides: Bedrock vs SageMaker | RAG Architecture | Prompt Engineering | Responsible AI | AI Services Decision Guide

Foundation Model Concepts

Pre-training

  • Training a model on massive datasets (trillions of tokens) to learn general language/world knowledge.
  • Extremely expensive and resource-intensive (millions of GPU hours).
  • Results in a base model with broad capabilities but no specific task alignment.
  • Common objectives: next-token prediction (GPT-style), masked language modeling (BERT-style).

Fine-tuning Techniques

  • Instruction Tuning — fine-tuning on instruction-response pairs to make the model follow instructions better.
  • RLHF (Reinforcement Learning from Human Feedback) — trains a reward model from human preferences, then uses RL (PPO) to optimize the language model against that reward. Used to align models with human values.
  • DPO (Direct Preference Optimization) — simpler alternative to RLHF that directly optimizes on preference pairs without a separate reward model. More stable training.
  • LoRA / QLoRA — Parameter-Efficient Fine-Tuning (PEFT) that freezes base model and trains small adapter layers. Reduces compute by 90%+.
  • Continued Pre-training — further pre-training on domain-specific data to teach the model new knowledge (e.g., medical, legal, financial).
  • Distillation — training a smaller “student” model to mimic a larger “teacher” model’s outputs. Reduces inference cost while retaining most capability.

RAG (Retrieval Augmented Generation)

  • Combines information retrieval with text generation to ground LLM responses in external knowledge.
  • How it works: Query → Retrieve relevant documents from knowledge base → Augment prompt with retrieved context → Generate response.
  • Benefits: Reduces hallucinations, enables up-to-date responses, no model retraining needed, source attribution.
  • Components: Document ingestion, chunking strategy, embedding model, vector database, retrieval algorithm, re-ranking.
  • AWS Implementation: Amazon Bedrock Knowledge Bases, Amazon Kendra (GenAI Index), OpenSearch vector search.

Prompt Engineering

  • Zero-shot — asking the model to perform a task without any examples. Relies on pre-trained knowledge.
  • Few-shot (In-Context Learning) — providing a few examples in the prompt to guide the model’s output format and behavior.
  • Chain-of-Thought (CoT) — asking the model to “think step by step” to improve reasoning on complex tasks.
  • System Prompts — instructions that define the model’s role, behavior, and constraints.
  • Prompt Templates — reusable prompt structures with placeholders for dynamic content.
  • Prompt Chaining — breaking complex tasks into sequential prompts where output of one feeds input of next.

Key Parameters & Concepts

  • Tokenization — splitting text into tokens (subwords/words). Models process tokens, not characters. Affects context limits and pricing.
  • Embeddings — dense vector representations of text/images in high-dimensional space. Semantically similar items have similar embeddings. Used for search, RAG, and clustering.
  • Temperature — controls randomness of output. Low (0-0.3) = deterministic/focused, High (0.7-1.0) = creative/diverse. 0 = greedy decoding.
  • Top-p (Nucleus Sampling) — considers only tokens whose cumulative probability exceeds p. Top-p 0.9 = considers top 90% probability mass.
  • Top-k — limits token selection to the k most likely next tokens.
  • Context Window — maximum number of tokens (input + output) the model can process at once. Ranges from 4K to 1M+ tokens in modern models.
  • Max Tokens — limits the length of generated output.
  • Stop Sequences — tokens that signal the model to stop generating.
  • Hallucination — when a model generates plausible-sounding but factually incorrect information.
  • Grounding — techniques to anchor model responses in factual data (RAG, tool use, citations).

Responsible AI

Core Principles

  • Fairness & Bias — ensuring models don’t discriminate based on protected attributes (race, gender, age). Types: selection bias, measurement bias, representation bias, confirmation bias.
  • Explainability — ability to understand and explain how/why a model made a specific prediction. Techniques: SHAP, LIME, attention visualization, feature importance.
  • Transparency — openly communicating model capabilities, limitations, and intended use cases to users.
  • Robustness — model performs reliably across different inputs, including adversarial examples and edge cases.
  • Privacy & Security — protecting training data, user inputs, and model outputs. Preventing data leakage and prompt injection.
  • Governance — organizational policies, processes, and controls for responsible AI development and deployment.
  • Safety — preventing harmful outputs including toxic content, misinformation, and dangerous instructions.

AWS Responsible AI Tools

  • AWS AI Service Cards — transparency documentation for AWS AI services covering intended use cases, limitations, responsible AI design choices, and deployment best practices.
  • Amazon Bedrock Guardrails — configurable safeguards for GenAI applications:
    • Content filters (hate, insults, sexual, violence, misconduct)
    • Denied topics (topic avoidance policies)
    • Word/phrase filters
    • Sensitive information filters (PII redaction)
    • Contextual grounding checks (hallucination detection)
    • Automated Reasoning Checks (logical verification)
  • SageMaker Clarify — detects bias in data and models, provides feature attributions for explainability (note: moving to maintenance July 2026 for new customers).
  • Model Cards — documentation that describes a model’s intended use, performance metrics, limitations, and ethical considerations. Supported in SageMaker Model Registry.
  • Human-in-the-Loop (HITL) — keeping humans involved in AI decision-making for high-stakes scenarios. AWS A2I (Augmented AI) provided review workflows (note: moving to maintenance July 2026 for new customers).
  • Amazon Bedrock Model Evaluation — automatic evaluation (accuracy, robustness, toxicity), human evaluation, and LLM-as-a-judge for quality assessment.

Bias Mitigation Strategies

  • Pre-processing: Balance training data, remove sensitive attributes, data augmentation.
  • In-processing: Regularization techniques, adversarial debiasing, fairness constraints during training.
  • Post-processing: Calibrate outputs, threshold adjustment, reject option classification.
  • Monitoring: Continuously track model performance across demographic groups in production.

Agentic AI

What are AI Agents?

  • AI systems that can autonomously plan, reason, and execute multi-step tasks to achieve goals.
  • Go beyond simple prompt-response by taking actions, using tools, and adapting based on outcomes.
  • Can operate for extended periods, making decisions and course-correcting without human intervention.

Key Concepts

  • Tool Use (Function Calling) — agents invoke external tools (APIs, databases, code execution) to gather information or perform actions.
  • Multi-step Reasoning — breaking complex problems into steps, executing sequentially with intermediate evaluations.
  • Orchestration — coordinating multiple agents or components to complete complex workflows. Patterns: sequential, parallel, routing, supervisor.
  • Memory — maintaining context across interactions:
    • Short-term memory (conversation context within a session)
    • Long-term memory (persistent knowledge across sessions)
    • Episodic memory (past experiences and outcomes)
  • Planning — decomposing goals into actionable sub-tasks, determining execution order, handling dependencies.
  • Reflection — agents evaluate their own outputs and self-correct errors before responding.
  • Model Context Protocol (MCP) — open standard for connecting AI agents with external tools and data sources.
  • Agent2Agent (A2A) — protocol for inter-agent communication and collaboration.

AWS Agentic AI Services

  • Amazon Bedrock Agents — create agents that can break down tasks, call APIs, and access knowledge bases (transitioning to Bedrock Agents Classic, July 2026).
  • Amazon Bedrock AgentCore (GA 2025/2026) — enterprise-grade infrastructure for deploying and operating AI agents at scale:
    • AgentCore Runtime — serverless, scalable environment to host agents
    • AgentCore Gateway — MCP-compatible tool connectivity
    • AgentCore Identity — per-agent identity and least-privilege access
    • AgentCore Observability — monitoring, tracing, and debugging
    • AgentCore Code Interpreter — secure sandboxed code execution
    • AgentCore Optimization — continuous quality evaluation and improvement
  • Amazon Nova Act — browser automation agent for web-based tasks.
  • AWS Step Functions — orchestrate multi-step agent workflows with state management.

AWS AI Service Stack

AWS organizes AI/ML services into three layers:

Layer 1: AI Infrastructure (Compute & Silicon)

Service/Chip Purpose Key Details
AWS Trainium Custom chip for ML training Trainium2 (4x perf vs gen1), Trainium3 (3nm, 4.4x vs Trn2, GA Dec 2025)
AWS Inferentia Custom chip for ML inference Inferentia2 (4x throughput, 10x lower latency vs gen1), Inf2 instances
EC2 UltraServers Multi-instance AI clusters Trn2 UltraServers (64 Trainium2 chips, NeuronLink interconnect), Trn3 UltraServers
AWS AI Factories On-premises AI infrastructure Deploy AI training/inference infrastructure in customer data centers
AWS Neuron SDK Software for Trainium/Inferentia Integrates with PyTorch, JAX, TensorFlow. Compiler, runtime, profiler
EC2 P5/P5e/P5en GPU instances (NVIDIA) H100/H200 GPUs for training and inference
EC2 G6/G6e GPU instances (NVIDIA) L4/L40S GPUs for inference and graphics
AWS Graviton Arm-based general compute Best price-performance for inference serving and general ML workloads
Amazon EFA Elastic Fabric Adapter Low-latency networking for distributed training across instances

Layer 2: ML Platform (SageMaker AI)

  • Amazon SageMaker AI (rebranded from SageMaker, late 2024) — end-to-end ML platform for building, training, and deploying models.
Component Purpose
SageMaker Unified Studio Single IDE for data, analytics, and ML/AI development (integrates Bedrock)
SageMaker Canvas No-code ML for business analysts — point-and-click model building
SageMaker HyperPod Managed clusters for large-scale distributed training with auto-recovery
SageMaker Pipelines CI/CD for ML — define, automate, and manage ML workflows
SageMaker Feature Store Centralized repository for ML features (online + offline store)
SageMaker MLflow Managed MLflow for experiment tracking, model versioning, deployment
SageMaker Model Registry Central catalog to version, manage, and deploy models with approval workflows
SageMaker JumpStart Model hub with 400+ pre-trained models, one-click deploy, fine-tuning
SageMaker Endpoints Real-time inference hosting (single model or multi-model endpoints)
SageMaker Training Managed training with built-in algorithms, distributed training, spot instances
SageMaker Processing Run data processing and evaluation jobs at scale
SageMaker Lakehouse Unified access to data lakes and warehouses for ML

Layer 3: AI Applications & Services

Amazon Bedrock (Generative AI Platform)

  • Amazon Bedrock — fully managed service for building GenAI applications with foundation models.
  • Model Providers: Amazon (Nova), Anthropic (Claude), Meta (Llama), Mistral, Cohere, AI21 Labs, OpenAI, Stability AI.
  • Key Capabilities:
    • Model Inference — Converse API, InvokeModel, streaming, batch inference, cross-region inference
    • Knowledge Bases — managed RAG with vector stores (OpenSearch, Aurora, Pinecone, etc.)
    • Managed Knowledge Base (2026) — fully managed RAG primitive (storage + retrieval + embeddings + re-ranking)
    • Agents — multi-step task execution with tool use (transitioning to AgentCore)
    • Guardrails — content filtering, topic avoidance, PII protection, grounding checks
    • Model Customization — fine-tuning, continued pre-training, distillation
    • Model Evaluation — automatic metrics, human evaluation, LLM-as-judge
    • Flows — visual workflow builder for chaining prompts, agents, and knowledge bases

Amazon Nova Models

  • Nova Micro — text-only, fastest, lowest cost (128K context). Ideal for classification, summarization.
  • Nova Lite — multimodal (text + image + video input), cost-effective (300K context).
  • Nova Pro — balanced multimodal, strong accuracy/speed/cost trade-off (300K context).
  • Nova Premier — most capable, complex reasoning, agentic workflows, teacher model (1M context).
  • Nova Canvas — image generation with editing controls and watermarking.
  • Nova Reel — video generation (1280×720, 24fps, up to 6 seconds).
  • Nova Sonic — speech-to-speech for real-time conversational AI.
  • Nova 2 (Dec 2025) — next generation with extended thinking (adjustable levels), 1M token context, built-in tools:
    • Nova 2 Lite — fast, cost-effective reasoning model
    • Nova 2 Pro — most intelligent, complex agentic tasks
    • Nova 2 Sonic — next-gen speech with async tool calling
    • Nova 2 Omni — unified multimodal I/O (text + image generation)
  • Nova Act — browser automation agent for web tasks.
  • Nova Forge — custom model building program (open training).

Amazon Q Developer & Q Business

  • Amazon Q Developer — AI-powered coding assistant (evolved from CodeWhisperer):
    • Code generation, completion, and inline suggestions (15+ languages)
    • Agentic coding (autonomous multi-step development)
    • Security vulnerability scanning
    • Code transformation and modernization (Java, .NET upgrades)
    • CLI integration (natural language → commands)
    • Debugging and troubleshooting with CloudWatch integration
  • Amazon Q Business — AI assistant for enterprise knowledge (connects to 40+ data sources):
    • Natural language answers from company data
    • Document summarization and content creation
    • Task automation with plugins
    • Access control respecting existing permissions (ACL-aware)
  • Amazon Q in Console — chat assistant in AWS Management Console for troubleshooting and guidance.
⚠️ Note (July 2026): Amazon Q Developer IDE plugins reaching end-of-support April 2027. Successor is Kiro — AWS’s agentic development environment. Amazon Q Business and Amazon Kendra entering maintenance mode for new customers July 30, 2026.

AWS AI/ML Application Services

Service Category Purpose
Amazon Comprehend NLP Sentiment analysis, entity recognition, key phrase extraction, language detection, topic modeling
Amazon Rekognition Computer Vision Object/face detection, content moderation, celebrity recognition, text in images, custom labels
Amazon Polly Speech Text-to-speech with neural voices (60+ voices, 30+ languages), SSML support
Amazon Transcribe Speech Speech-to-text (ASR), real-time and batch, custom vocabularies, speaker identification
Amazon Translate Language Neural machine translation (75+ languages), real-time and batch, custom terminology
Amazon Textract Document AI OCR + intelligent document processing, extracts text, tables, forms, and queries from documents
Amazon Lex Conversational AI Build chatbots and voice bots with automatic speech recognition and NLU
Amazon Kendra Search Enterprise search with NLP, semantic understanding, GenAI index for RAG ⚠️ Maintenance mode July 2026
Amazon Personalize Recommendations Real-time personalization and recommendations (same tech as Amazon.com)
Amazon Forecast Time Series Time series forecasting using ML (closed to new customers since 2024)
Amazon HealthScribe Healthcare Generate clinical documentation from patient-clinician conversations
Amazon Bedrock AgentCore Agentic AI Deploy, manage, and optimize AI agents at scale (GA 2025/2026)

Decision Matrix: Use Case → Recommended Service

Use Case Recommended Service Why
Build GenAI apps with FMs (no ML expertise) Amazon Bedrock Serverless, multi-model, fully managed
Custom model training from scratch SageMaker AI + Trainium Full control over training, data, and infrastructure
Enterprise Q&A over company documents Amazon Q Business / Bedrock Knowledge Bases Connects to 40+ data sources, ACL-aware
AI coding assistant Amazon Q Developer / Kiro Inline completions, security scanning, agentic coding
Build and deploy AI agents Bedrock AgentCore Serverless runtime, MCP tools, identity, observability
Chatbot / virtual assistant Amazon Lex + Bedrock Lex for structure, Bedrock for natural responses
Document processing (forms, invoices) Amazon Textract Extracts structured data from documents at scale
Content moderation (images/video) Amazon Rekognition Pre-built moderation labels, custom labels for specifics
Sentiment analysis on customer feedback Amazon Comprehend Pre-built NLP models, no training needed
Real-time product recommendations Amazon Personalize Same ML tech as Amazon.com, real-time updates
Transcribe meetings/calls Amazon Transcribe Real-time ASR, speaker diarization, custom vocab
Generate speech from text Amazon Polly Neural TTS, SSML support, multiple voices
Translate content at scale Amazon Translate 75+ languages, real-time, custom terminology
No-code ML for business users SageMaker Canvas Point-and-click, AutoML, visual interface
Fine-tune FMs on proprietary data Bedrock Custom Models / SageMaker JumpStart Bedrock for serverless; SageMaker for full control
Prevent harmful GenAI outputs Amazon Bedrock Guardrails Content filters, PII, grounding checks, topic avoidance
Cost-effective GenAI inference at scale Bedrock + Nova models (or Inferentia2/Trainium) Nova = lowest cost in class; custom silicon for self-hosted
Clinical documentation from conversations Amazon HealthScribe Purpose-built for healthcare, HIPAA eligible

Quick Reference: All AWS AI/ML Services

Service One-Liner
Amazon Bedrock Fully managed GenAI platform with multi-provider foundation models
Amazon Bedrock AgentCore Enterprise infrastructure for deploying and operating AI agents at scale
Amazon Nova Amazon’s family of foundation models (text, multimodal, speech, image, video)
Amazon Q Developer AI coding assistant with code generation, security scanning, and transformation
Amazon Q Business Enterprise AI assistant for Q&A and task automation over company data
Amazon SageMaker AI End-to-end ML platform for building, training, and deploying custom models
Amazon Comprehend NLP service for sentiment, entities, key phrases, language detection
Amazon Rekognition Computer vision for object/face detection, moderation, and custom labels
Amazon Polly Text-to-speech with neural and standard voices
Amazon Transcribe Automatic speech recognition (speech-to-text)
Amazon Translate Neural machine translation for 75+ languages
Amazon Textract Extract text, tables, and forms from documents (OCR+)
Amazon Lex Build conversational chatbots and voice bots
Amazon Kendra Intelligent enterprise search with NLP and GenAI index
Amazon Personalize Real-time ML-powered personalization and recommendations
Amazon HealthScribe Generate clinical notes from patient-clinician conversations
AWS Trainium Custom AI chip optimized for training (Trn2, Trn3 instances)
AWS Inferentia Custom AI chip optimized for inference (Inf2 instances)
AWS Neuron SDK SDK for running ML workloads on Trainium and Inferentia chips
Amazon SageMaker Canvas No-code ML model building for business analysts
Amazon SageMaker HyperPod Managed clusters for distributed training with auto fault recovery

Exam Tips

AIF-C01 — AWS Certified AI Practitioner

  • Format: 65 questions, 90 minutes, 700/1000 passing score.
  • Domains:
    • Domain 1: Fundamentals of AI and ML (20%)
    • Domain 2: Fundamentals of Generative AI (24%)
    • Domain 3: Applications of Foundation Models (28%) — largest domain, most technical
    • Domain 4: Guidelines for Responsible AI (14%)
    • Domain 5: Security, Compliance, and Governance for AI Solutions (14%)
  • Key Focus Areas:
    • Domains 2+3 = 52% of exam — master Bedrock, RAG, prompt engineering, fine-tuning
    • Know the difference between AI vs ML vs DL vs GenAI
    • Understand when to use Bedrock vs SageMaker
    • RAG architecture and when to use it vs fine-tuning
    • Responsible AI principles and Bedrock Guardrails
    • Temperature, top-p effects on output
    • Know all AWS AI services at a high level (what each does)

AIP-C01 — AWS Certified Generative AI Developer – Professional

  • Format: 85 questions, 180 minutes, 750/1000 passing score.
  • Domains:
    • Domain 1: FM Selection and Integration (26%)
    • Domain 2: Data Management and Optimization (22%)
    • Domain 3: Model Performance and Compliance (31%) — largest domain
    • Domain 4: Security and Governance (21%)
  • Key Focus Areas:
    • Deep hands-on knowledge of Bedrock APIs, agents, knowledge bases, guardrails
    • RAG implementation details (chunking strategies, embedding models, vector stores)
    • Model customization (when fine-tuning vs RAG vs prompt engineering)
    • Agentic AI patterns (tool use, multi-step, AgentCore)
    • SageMaker for custom training and deployment
    • Model evaluation and monitoring in production
    • Security: data encryption, VPC endpoints, IAM for Bedrock, prompt injection mitigation
    • Cost optimization (model selection, batch inference, provisioned throughput)

Common Exam Scenarios

  • “Least operational overhead” → Bedrock (serverless) over SageMaker (managed infrastructure)
  • “Custom model with proprietary data” → Fine-tuning on Bedrock or SageMaker depending on control needed
  • “Reduce hallucinations” → RAG with Knowledge Bases + Guardrails grounding checks
  • “Enterprise search over internal docs” → Amazon Q Business or Bedrock Knowledge Bases
  • “Control AI outputs for safety” → Bedrock Guardrails
  • “Lowest cost inference” → Nova Micro (text) or Nova Lite (multimodal) on Bedrock
  • “Deploy agents in production” → Bedrock AgentCore (serverless, scalable, observable)
  • “Train trillion-parameter model” → Trainium3 UltraServers + SageMaker HyperPod

Practice Questions

Question 1 (AIF-C01)

A company wants to reduce hallucinations in their generative AI application that answers customer questions about company policies. The application uses Amazon Bedrock. What is the MOST effective approach?

  1. Increase the model temperature to generate more diverse responses
  2. Implement Retrieval Augmented Generation (RAG) with Amazon Bedrock Knowledge Bases
  3. Fine-tune the foundation model on company documents
  4. Switch to a larger foundation model
Show Answer

Answer: B – RAG grounds responses in actual company documents, directly reducing hallucinations. Fine-tuning (C) teaches style/format but doesn’t guarantee factual accuracy for specific documents. Higher temperature (A) increases randomness. Larger models (D) don’t inherently reduce hallucinations.

Question 2 (AIF-C01)

Which combination of techniques helps ensure responsible AI in a generative AI application? (Select TWO)

  1. Increase the context window size
  2. Configure Amazon Bedrock Guardrails with content filters and denied topics
  3. Use the lowest-cost foundation model available
  4. Implement human review workflows for high-stakes decisions
  5. Maximize the temperature parameter for creative outputs
Show Answer

Answer: B, D – Bedrock Guardrails (B) provides configurable safety controls to filter harmful content. Human-in-the-loop (D) ensures human oversight for critical decisions. Context window size (A), model cost (C), and temperature (E) are not responsible AI techniques.

Question 3 (AIP-C01)

A developer is building an AI agent that needs to autonomously execute multi-step workflows, call external APIs, and maintain state across interactions. The solution must be production-grade with monitoring and minimal infrastructure management. Which AWS service should they use?

  1. Amazon Lex with Lambda fulfillment functions
  2. Amazon Bedrock AgentCore with AgentCore Runtime and Observability
  3. AWS Step Functions with SageMaker endpoints
  4. Amazon Q Business with custom plugins
Show Answer

Answer: B – Bedrock AgentCore provides serverless runtime for agents, MCP-compatible tool connectivity (Gateway), built-in observability, and identity management — purpose-built for production AI agents. Lex (A) is for chatbots, not autonomous agents. Step Functions (C) requires more infrastructure management. Q Business (D) is for enterprise knowledge, not custom agent workflows.

Question 4 (AIP-C01)

A team needs to fine-tune a foundation model on their proprietary dataset with minimal compute cost. The dataset contains 10,000 instruction-response pairs. Which approach provides the BEST balance of performance improvement and cost?

  1. Full fine-tuning of the entire model on Amazon SageMaker with P5 GPU instances
  2. Continued pre-training on Amazon Bedrock with the full dataset
  3. Parameter-efficient fine-tuning (LoRA) through Amazon Bedrock custom models
  4. Distilling the model into a smaller variant using Nova Premier as teacher
Show Answer

Answer: C – LoRA fine-tuning on Bedrock trains only small adapter layers (reduces compute by 90%+) while the base model stays frozen. It’s ideal for instruction-tuning with limited data. Full fine-tuning (A) is expensive. Continued pre-training (B) is for teaching new knowledge, not task alignment. Distillation (D) creates a smaller model but doesn’t directly fine-tune on task data.

Question 5 (AIF-C01 / AIP-C01)

A company wants to deploy a generative AI solution with the following requirements: lowest possible latency for text summarization, minimal cost, and no infrastructure management. Which combination should they choose?

  1. Amazon Nova Premier on Bedrock with provisioned throughput
  2. Amazon Nova Micro on Bedrock with on-demand pricing
  3. Claude 3 Opus on Bedrock with batch inference
  4. Self-hosted Llama model on SageMaker with Inferentia2 instances
Show Answer

Answer: B – Nova Micro is the fastest text-only model (200+ tokens/sec), lowest cost, and Bedrock provides serverless (no infrastructure). Premier (A) is more capable but slower and costlier. Batch (C) has high latency. Self-hosted (D) requires infrastructure management.

Frequently Asked Questions

What is the difference between AI, ML, and Generative AI?

AI is the broadest category — machines performing tasks that typically require human intelligence. ML is a subset that learns from data without explicit programming. Generative AI is a subset of ML that creates new content (text, images, code) using foundation models trained on vast datasets.

What is the difference between Amazon Bedrock and SageMaker?

Bedrock provides access to pre-built foundation models for generative AI applications without ML expertise. SageMaker is a full ML platform for building, training, and deploying custom models from scratch. Use Bedrock for gen AI apps; SageMaker when you need complete control over model training.

What AWS certifications cover AI and Generative AI?

AWS offers two AI-focused certifications: AIF-C01 (AI Practitioner) for foundational knowledge of AI/ML/Gen AI concepts and AWS services, and AIP-C01 (AI Professional) for practitioners building and deploying Gen AI solutions. Both require knowledge of Bedrock, SageMaker, and responsible AI.

Detailed Guides

Exam Prep: AWS AI Professional (AIP-C01) Exam Learning Path

References

Amazon Nova Models – Capabilities, Pricing & Use Cases Compared

~ Last updated on : ~ Kiro Agent

Amazon Nova Models Overview

Amazon Nova is AWS’s family of foundation models (FMs) available exclusively through Amazon Bedrock. Launched at AWS re:Invent 2024 and significantly expanded at re:Invent 2025, the Nova family spans text generation, multimodal understanding, image generation, video generation, speech-to-speech conversation, browser automation, and custom model building.

Nova models are designed to deliver frontier intelligence at industry-leading price performance, making them a compelling choice for enterprises looking to reduce costs while maintaining high-quality AI outputs.

Amazon Nova Model Family

The Nova family is organized into several categories:

  • Understanding Models – Process text, images, video, and/or speech to generate text (Nova Micro, Nova Lite, Nova Pro, Nova Premier, Nova 2 Lite, Nova 2 Pro)
  • Creative Content Generation – Generate images or video from text/image inputs (Nova Canvas, Nova Reel)
  • Speech Models – Real-time bidirectional voice conversation (Nova Sonic, Nova 2 Sonic)
  • Multimodal Generation – Unified input and output across modalities (Nova 2 Omni)
  • Agentic – Browser-based UI automation (Nova Act)
  • Custom Model Building – Build your own frontier model variants (Nova Forge)

First Generation Nova Models (re:Invent 2024)

Amazon Nova Micro

  • Text-only model delivering the lowest latency responses in the Nova family at very low cost.
  • Context window: 128K tokens
  • Input: Text only
  • Output: Text only
  • Optimized for speed and cost — excels at text summarization, translation, content classification, interactive chat, brainstorming, and simple mathematical reasoning and coding.
  • Supports customization via fine-tuning and model distillation.
  • Pricing: ~$0.035/1M input tokens, ~$0.14/1M output tokens (lowest in the Nova family)

Amazon Nova Lite

  • Very low-cost multimodal model that is lightning fast for processing image, video, and text inputs.
  • Context window: 300K tokens
  • Input: Text, images, video (up to 30 minutes)
  • Output: Text
  • Handles real-time customer interactions, document analysis, and visual question-answering with high accuracy.
  • Supports text and multimodal fine-tuning and model distillation.
  • Pricing: ~$0.06/1M input tokens, ~$0.24/1M output tokens

Amazon Nova Pro

  • Highly capable multimodal model with the best combination of accuracy, speed, and cost for a wide range of tasks.
  • Context window: 300K tokens
  • Input: Text, images, video
  • Output: Text
  • Sets standards in multimodal intelligence and agentic workflows requiring API/tool calling.
  • Excels at visual question answering, video understanding, financial document analysis, and processing code bases with over 15,000 lines.
  • Serves as a teacher model for distilling custom variants of Nova Micro and Lite.
  • Pricing: ~$0.80/1M input tokens, ~$3.20/1M output tokens

Amazon Nova Premier

  • Most capable multimodal model for complex reasoning tasks and the best teacher for distilling custom models.
  • Context window: 1M tokens
  • Input: Text, images, video
  • Output: Text
  • Designed for the most demanding enterprise workloads requiring highest accuracy.
  • Best suited for complex multi-step reasoning, research, and as a distillation teacher.

Amazon Nova Canvas

  • State-of-the-art image generation model producing studio-quality images.
  • Input: Text prompts, reference images
  • Output: Images (up to 2048×2048)
  • Features precise control over style and content with rich editing capabilities:
    • Inpainting and outpainting
    • Background removal
    • Color conditioning
    • Subject-consistent generation
  • Includes built-in watermarking for responsible AI use.
  • Pricing: Per-image pricing, varies by resolution

Amazon Nova Reel

  • State-of-the-art video generation model for producing short videos.
  • Input: Text prompts, reference images
  • Output: Video (1280×720, 24fps, 6-second clips)
  • Supports camera control actions (zoom, pan, tilt, dolly) and visual style guidance.
  • Ideal for marketing, advertising, social media content, and entertainment.
  • Uses asynchronous API (StartAsyncInvoke) as video generation takes several minutes.
  • Includes digital watermarking on all generated videos.

Amazon Nova Sonic

  • Speech-to-speech foundation model enabling real-time, human-like voice conversations with low latency.
  • Input: Speech (audio), text
  • Output: Speech (audio), text
  • Unifies speech understanding and generation into a single model architecture.
  • Supports natural turn-taking, interruption handling, and expressive voices (masculine and feminine).
  • Multilingual support: English, Spanish, French, Italian, and German.
  • Integrates with Amazon Connect, telephony providers (Vonage, Twilio, AudioCodes), and conversational AI frameworks (LiveKit, Pipecat).
  • Uses bidirectional streaming API for real-time interaction.

Second Generation Nova Models (re:Invent 2025)

At AWS re:Invent 2025, Amazon announced the Nova 2 family with four new models, plus Nova Forge and Nova Act.

Amazon Nova 2 Lite

  • Fast, cost-effective reasoning model for everyday workloads.
  • Context window: 1M tokens (1 million)
  • Input: Text, images, video, documents
  • Output: Text (up to 65,536 tokens)
  • Features adjustable thinking effort — developers can control how much step-by-step reasoning the model performs before responding, balancing intelligence depth with speed and cost.
  • Built-in tools:
    • Web grounding — searches the web for current information with citations
    • Code interpreter — runs code directly within the workflow
    • Remote MCP tool support
  • Excels in document processing, video information extraction, code generation, grounded answers, and multi-step agentic workflows.
  • Competitive benchmarks: Equal or better on 13/15 benchmarks vs Claude Haiku 4.5, 11/17 vs GPT-5 Mini, 14/18 vs Gemini Flash 2.5.
  • Supports customization via Nova Forge.
  • Pricing: ~$0.30–0.33/1M input tokens, ~$2.50–2.75/1M output tokens

Amazon Nova 2 Pro

  • Amazon’s most intelligent reasoning model for highly complex tasks.
  • Context window: 1M tokens
  • Input: Text, images, video, speech
  • Output: Text (up to 65,536 tokens)
  • Ideal for agentic coding, long-range planning, and sophisticated problem-solving where highest accuracy is essential.
  • Serves as a teacher model for knowledge distillation into smaller, more efficient student models.
  • Built-in web grounding and code execution capabilities.
  • Competitive benchmarks: Equal or better on 10/16 vs Claude Sonnet 4.5, 8/16 vs GPT-5.1, 15/19 vs Gemini 2.5 Pro.
  • Strengths: Multi-document analysis, video reasoning, complex instructions, advanced math, agentic and software engineering tasks.

Amazon Nova 2 Sonic

  • Next-generation speech-to-speech model for real-time conversational AI.
  • Context window: 1M tokens
  • Input: Speech, text
  • Output: Speech, text
  • Key improvements over Nova Sonic:
    • Expanded multilingual support with native expressivity
    • Higher accuracy and more natural voices
    • Cross-modal interaction — seamless switching between voice and text in the same session
    • Asynchronous tool calling — handles tasks in the background without interrupting conversation
    • 1M token context window for sustained interactions
  • Integrates with Amazon Connect, Vonage, Twilio, AudioCodes, LiveKit, and Pipecat.
  • Industry-leading price performance vs OpenAI gpt-realtime and Gemini 2.5 Flash realtime APIs.

Amazon Nova 2 Omni

  • Unified multimodal reasoning AND generation model — an industry first.
  • Context window: 1M tokens
  • Input: Text, images, video, speech, documents
  • Output: Text AND images
  • Can process up to 750,000 words, hours of audio, long videos, and hundred-page documents simultaneously.
  • Eliminates the need to connect multiple specialized models — handles understanding and generation in one workflow.
  • Supports 200+ languages for text processing, 10 languages for speech input.
  • Image generation capabilities include character consistency, text rendering within images, and object/background modification.
  • Use case example: Marketing teams can analyze product details across all formats and generate complete campaigns (headlines, copy, social posts, visuals) in one workflow.
  • Status: Preview (as of December 2025)

Amazon Nova Act

  • AWS service for building and deploying highly reliable AI agents that automate browser-based UI workflows.
  • Powered by a custom Nova 2 Lite model trained through reinforcement learning on hundreds of simulated web environments.
  • Achieves 90% reliability on early customer workflows — outperforms competing models on relevant benchmarks.
  • Key capabilities:
    • No-code playground for prototyping agents with natural language prompts
    • IDE integration (VS Code) for refinement
    • Deploy to AWS with comprehensive management tools and monitoring via Nova Act console
    • What you build locally scales in production
  • Use cases:
    • Updating data in CRM systems
    • Testing website functionality (QA automation)
    • Submitting health insurance claims
    • Reconciling payments and coordinating shipments
    • End-to-end testing across platforms
  • Customer results:
    • Hertz: 5x faster software delivery, eliminated QA bottleneck
    • 1Password: Automated logins across hundreds of websites
    • Sola Systems: Hundreds of thousands of automated workflows per month
  • Status: Generally Available (December 2025)

Amazon Nova Forge

  • First-of-its-kind service for organizations to build their own frontier AI model variants (“Novellas”).
  • Pioneers “open training” — gives exclusive access to pre-trained, mid-trained, and post-trained Nova model checkpoints.
  • Customers can mix proprietary data with Amazon Nova-curated datasets at every stage of model training.
  • Solves the three compromises of traditional model customization:
    • Surface-level customization of proprietary models
    • Open-weights models losing capabilities during continued training
    • Building from scratch at enormous expense
  • Key capabilities:
    • Open training checkpoints — blend proprietary data at pre-training, mid-training, and post-training stages
    • Reinforcement learning “gyms” — synthetic environments for training models on simulated real-world scenarios
    • Synthetic data-based distillation — create smaller, faster models that maintain intelligence
    • Responsible AI toolkit — implement safety controls
  • Currently available with Nova 2 Lite; early access to Nova 2 Pro and Nova 2 Omni for Forge customers.
  • Custom models deploy on Amazon Bedrock with enterprise-grade security, scalability, and data privacy.
  • Customers: Booking.com, Reddit, Sony, Cosine AI, Nimbus Therapeutics, Nomura Research Institute, OpenBabylon.

Amazon Nova Model Comparison Table

Model Input Types Output Types Context Window Latency Price Tier Best For
Nova Micro Text Text 128K Lowest $ (Lowest) Chat, summarization, classification
Nova Lite Text, Image, Video Text 300K Very Low $ Multimodal Q&A, document analysis
Nova Pro Text, Image, Video Text 300K Low $$ Agentic workflows, complex analysis
Nova Premier Text, Image, Video Text 1M Medium $$$ Complex reasoning, model distillation
Nova Canvas Text, Image Image N/A Medium Per image Image generation, editing
Nova Reel Text, Image Video N/A High (async) Per video Video content creation
Nova Sonic Speech, Text Speech, Text N/A Real-time $$ Voice assistants, IVR
Nova 2 Lite Text, Image, Video, Docs Text 1M Low $$ Reasoning, coding, agentic tasks
Nova 2 Pro Text, Image, Video, Speech Text 1M Medium $$$ Complex reasoning, agentic coding
Nova 2 Sonic Speech, Text Speech, Text 1M Real-time $$ Conversational AI, async tool use
Nova 2 Omni Text, Image, Video, Speech Text, Image 1M Medium $$$ Unified multimodal workflows
Nova Act Natural language prompts Browser actions N/A Variable Per workflow UI automation, QA testing
Nova Forge Training data + checkpoints Custom model N/A N/A (training) Custom Building domain-specific models

When to Use Each Nova Model

Scenario Recommended Model Why
High-volume chatbot with lowest cost Nova Micro / Nova 2 Lite Lowest per-token cost, fast response times
Document/image analysis at scale Nova 2 Lite 1M context, multimodal, cost-effective reasoning
Complex multi-step agentic coding Nova 2 Pro Highest intelligence, built-in tools
Building custom domain model Nova Forge + Nova 2 Lite/Pro Open training with proprietary data
Marketing image generation Nova Canvas Studio-quality with precise controls
Social media video creation Nova Reel Professional video from text/image prompts
Voice-based customer service Nova 2 Sonic Real-time, natural conversation with tool use
Unified content generation pipeline Nova 2 Omni Single model for text + image generation from any input
Browser-based QA testing Nova Act 90% reliability, production-ready UI automation
Knowledge distillation Nova Premier / Nova 2 Pro Best teacher models for creating efficient custom variants

Amazon Nova vs Competitors

Capability Amazon Nova Anthropic Claude OpenAI GPT Meta Llama
Model range Micro to Premier + specialized (Canvas, Reel, Sonic, Act) Haiku, Sonnet, Opus GPT-4o mini, GPT-4o, GPT-5 8B, 70B, 405B (open weights)
Multimodal input Text, image, video, speech, documents Text, image, documents Text, image, audio Text, image
Image generation Nova Canvas, Nova 2 Omni No DALL-E (separate) No
Video generation Nova Reel No Sora (separate) No
Speech-to-speech Nova Sonic / Nova 2 Sonic No GPT Realtime API No
Context window Up to 1M tokens 200K tokens 128K–1M tokens 128K tokens
Browser automation Nova Act Computer Use (beta) Operator No
Custom model building Nova Forge (open training) No Fine-tuning only Full open weights
Price performance Industry-leading (75% less than comparable models) Premium pricing Premium pricing Free weights (self-host costs)
AWS integration Native (Bedrock, Connect, S3) Via Bedrock Via Bedrock (select models) Via Bedrock

Key Differentiators for Amazon Nova

  • Breadth of modalities: No other single model family covers text, image, video, speech generation + browser automation + custom model building.
  • Price performance: Nova models consistently offer 3-5x better cost efficiency than comparable alternatives.
  • Nova Forge uniqueness: Only platform offering “open training” with access to model checkpoints at pre/mid/post-training stages.
  • Native AWS ecosystem: Deep integration with Bedrock, Knowledge Bases, Agents, Guardrails, Connect, and S3.
  • Built-in safety: All models include safety controls; creative models include watermarking.

Amazon Bedrock Integration

All Nova models are accessed through Amazon Bedrock, providing:

  • Unified API: Single Converse API for all text/multimodal models; async invoke for creative models.
  • Amazon Bedrock Knowledge Bases: Enhance Nova models with proprietary information via RAG.
  • Amazon Bedrock Agents: Build multi-step agentic workflows with Nova models.
  • Amazon Bedrock Guardrails: Apply content filters, PII detection, and grounding checks.
  • Cross-Region Inference: Available for Nova Micro, Lite, Pro, and Nova 2 models across multiple regions.
  • Intelligent Prompt Routing: Automatically route between Nova Pro and Nova Lite based on prompt complexity (up to 30% cost savings).
  • Model Customization: Fine-tuning, continued pre-training, and distillation support.
  • Batch Inference: 50% lower price for high-volume offline workloads.
  • Provisioned Throughput: Guaranteed capacity for production workloads.

Bedrock Model IDs

  • amazon.nova-micro-v1:0 – Nova Micro
  • amazon.nova-lite-v1:0 – Nova Lite
  • amazon.nova-pro-v1:0 – Nova Pro
  • amazon.nova-premier-v1:0 – Nova Premier
  • amazon.nova-canvas-v1:0 – Nova Canvas
  • amazon.nova-reel-v1:0 – Nova Reel
  • amazon.nova-sonic-v1:0 – Nova Sonic
  • amazon.nova-2-lite-v1:0 – Nova 2 Lite
  • amazon.nova-2-pro-v1:0 – Nova 2 Pro

AWS Certification – AIF-C01 Relevance

Amazon Nova models are highly relevant for the AWS Certified AI Practitioner (AIF-C01) exam:

  • Domain 1 – Fundamentals of AI and ML: Understanding foundation model types, model selection criteria, multimodal capabilities.
  • Domain 2 – Fundamentals of Generative AI: Foundation model families, model selection based on use case requirements (cost, latency, accuracy), understanding of text/image/video/speech generation.
  • Domain 3 – Applications of Foundation Models: RAG integration, agentic workflows, model customization (fine-tuning vs distillation vs continued pre-training vs open training).
  • Domain 4 – Guidelines for Responsible AI: Built-in safety controls, watermarking, Bedrock Guardrails integration.

Key exam concepts related to Nova:

  • Difference between understanding models and generative models
  • Model selection based on latency, cost, and capability requirements
  • Customization methods: fine-tuning, distillation, continued pre-training, Nova Forge open training
  • Responsible AI: watermarking, safety controls, content moderation
  • Bedrock features: Guardrails, Knowledge Bases, Agents, cross-region inference
  • Agentic AI capabilities and Nova Act for UI automation

Amazon Nova Models – Practice Questions

1. A company needs to build a high-volume customer service chatbot that must respond to text queries with the lowest possible latency and minimal cost. Which Amazon Nova model should they choose?

  1. Amazon Nova Pro
  2. Amazon Nova Micro
  3. Amazon Nova 2 Lite
  4. Amazon Nova Premier
Show Answer

Answer: B –

Explanation: Amazon Nova Micro is a text-only model designed specifically for the lowest latency responses at the lowest cost in the Nova family. It is optimized for tasks like interactive chat, text summarization, and content classification. While Nova 2 Lite also offers good performance, Nova Micro provides the absolute lowest latency for text-only use cases.

2. A media company wants to use a single AI model to analyze product videos, customer testimonials (audio), brand guidelines (documents), and product images — then generate marketing copy AND promotional images in one unified workflow. Which Amazon Nova model enables this?

  1. Amazon Nova Pro
  2. Amazon Nova Canvas
  3. Amazon Nova 2 Omni
  4. Amazon Nova 2 Pro + Nova Canvas (pipeline)
Show Answer

Answer: C –

Explanation: Amazon Nova 2 Omni is the industry’s first unified multimodal reasoning and generation model that can process text, images, video, and speech inputs while generating BOTH text AND images. This eliminates the need to chain multiple specialized models. Nova Pro and Nova 2 Pro only output text, and Nova Canvas only generates images from text/image prompts without the multimodal understanding capability.

3. An organization wants to build a custom AI model that deeply integrates their proprietary medical research data while maintaining frontier-level reasoning capabilities. They need access to model training checkpoints to mix their data at multiple training stages. Which Amazon service should they use?

  1. Amazon Bedrock Fine-tuning
  2. Amazon SageMaker Training
  3. Amazon Nova Forge
  4. Amazon Nova 2 Pro with RAG
Show Answer

Answer: C –

Explanation: Amazon Nova Forge is the only service that provides “open training” — giving organizations access to pre-trained, mid-trained, and post-trained Nova model checkpoints so they can blend their proprietary data with Nova-curated datasets at every stage of model training. Standard fine-tuning only adjusts a model’s behavior at the post-training stage. RAG augments responses but doesn’t deeply integrate knowledge into model weights.

4. A company wants to automate end-to-end testing of their web application, including navigating pages, filling forms, and verifying UI elements. They need a solution that achieves high reliability and can scale in production on AWS. Which Amazon Nova offering should they use?

  1. Amazon Nova 2 Pro with Bedrock Agents
  2. Amazon Nova Act
  3. Amazon Nova 2 Lite with custom prompts
  4. Amazon Nova Forge
Show Answer

Answer: B –

Explanation: Amazon Nova Act is specifically designed for building and deploying AI agents that automate browser-based UI workflows. It achieves 90% reliability through reinforcement learning training on hundreds of simulated web environments. It provides a no-code playground for prototyping, IDE integration for development, and production deployment on AWS with monitoring. Hertz used Nova Act to accelerate software delivery 5x by automating end-to-end testing.

5. A developer is building a voice-based AI assistant that needs to handle real-time conversations, switch between voice and text seamlessly, execute background tasks (like booking flights) without interrupting the conversation, and maintain context over long interactions. Which Amazon Nova model is best suited? (Select TWO capabilities that make it ideal)

  1. Amazon Nova Sonic — because it supports bidirectional streaming
  2. Amazon Nova 2 Sonic — because it supports asynchronous tool calling and 1M token context window
  3. Amazon Nova 2 Lite — because it has web grounding and code interpreter
  4. Amazon Nova 2 Omni — because it generates text and images
  5. Amazon Nova 2 Sonic — because it supports cross-modal interaction (voice/text switching)
Show Answer

Answer: B, E

Explanation: Amazon Nova 2 Sonic is the ideal model for this use case. It features: (1) asynchronous tool calling that lets users continue natural conversations while background tasks complete, (2) cross-modal interaction allowing seamless switching between voice and text, and (3) a 1M token context window for sustained interactions. While Nova Sonic (v1) supports real-time streaming, it lacks the async tool calling, cross-modal switching, and extended context window that Nova 2 Sonic provides.

Frequently Asked Questions

What are Amazon Nova models?

Amazon Nova is AWS’s own family of foundation models available through Bedrock. It includes text models (Micro, Lite, Pro, Premier), creative models (Canvas for images, Reel for video, Sonic for speech), and agent models (Act for browser automation, Forge for custom training).

Which Nova model should I use?

Use Nova Micro for fast text tasks (lowest cost). Nova Lite for multimodal with budget constraints. Nova Pro for balanced performance. Nova Premier for complex reasoning. Nova 2 Lite for extended thinking with 1M token context. Nova Act for browser-based agent automation.

How does Nova compare to Claude and GPT?

Nova Pro offers competitive performance at lower cost than Claude 3.5 Sonnet or GPT-4o for many tasks. Nova’s key advantages are native AWS integration, lower pricing, and specialized models (Canvas, Reel, Sonic) that competitors don’t offer as integrated options.

References