In this walkthrough, we explore the Discover menu in Amazon Bedrock. You'll learn how to navigate the Overview dashboard, dive deeper into the Model Catalog — filtering models by collection (Serverless vs. Marketplace), provider, and modality (Text, Image, Embedding, Speech, Video). You'll learn how to read model detail pages including pricing, specifications, and Model IDs.

We also cover the API Keys section, including the difference between short-term and long-term keys and when to use each.

In this walkthrough, we explore the Test menu. After a brief mention of the Chat/Text and Image/Video playgrounds (covered in previous lectures), we focus on two practical tools: Watermark detection — how to verify whether an image was generated by Amazon's Titan Image Generator or Nova Canvas models using invisible watermark analysis — and the Tokenizer — how to calculate token counts for your prompts before running inference, helping you estimate costs and stay within model context limits.

In this lecture, we explore the first two capabilities in the Infer menu. You'll learn how Cross-region inference uses inference profiles to automatically route requests across AWS Regions for better throughput and resilience. Then we cover Batch inference — how to process large volumes of requests asynchronously using S3 input/output for cost-effective, large-scale processing.

We continue with Provisioned Throughput — how to reserve dedicated model capacity for guaranteed performance and predictable costs in production workloads. Then we cover Custom model on-demand, which lets you deploy fine-tuned or imported models for pay-per-request inference without maintaining always-on infrastructure.

This lecture covers Custom models and the three customization techniques available in Bedrock. You'll learn how Reinforcement fine-tuning uses reward functions instead of labeled data, how Supervised fine-tuning trains models with input-output pairs, and how Distillation creates smaller, faster models from larger teacher models — up to 500% faster at up to 75% lower cost.

We explore Prompt router models (Intelligent Prompt Routing) — how to automatically route requests between a larger and smaller model based on prompt complexity, reducing costs by up to 30% without sacrificing quality. You'll learn how default and custom routers work, how to set quality thresholds, and how the fallback model serves as your cost-saving baseline.

In this lecture, we cover Imported models — how to bring models you've fine-tuned in SageMaker or other environments into Bedrock using Custom Model Import. Then we explore Marketplace model deployments — subscribing to and deploying specialized models from 100+ providers through Bedrock's unified API.

We begin exploring the Build menu with Agents and Flows. You'll learn how Agents orchestrate foundation models, APIs, and data sources to complete tasks autonomously, including multi-agent collaboration. Then we cover Flows — the visual drag-and-drop builder for creating AI workflows using prompt nodes, knowledge base nodes, condition logic, and Lambda functions.

This lecture covers Knowledge Bases (RAG), Automated Reasoning, and Guardrails. You'll learn how Knowledge Bases ground model responses in your proprietary data using Retrieval Augmented Generation. Then we explore Automated Reasoning for mathematically validating responses against business rules, and Guardrails with its six safeguard policies for content filtering, PII protection, and hallucination prevention.

We continue with Prompt Management — centralized prompt creation, versioning, testing with side-by-side comparison, and optimization. Then we cover Data Automation for extracting structured insights from documents, images, video, and audio using blueprints and custom field extraction.

In this lecture, we explore AgentCore — the platform for deploying AI agents at production scale. You'll learn about its modular services: Runtime for auto-scaling, Gateway for API-to-tool transformation, Memory for persistent and episodic storage, Identity for authentication, Browser Tool, and Code Interpreter. We also cover Policy enforcement using Cedar and the built-in Evaluations framework.

We explore the Evaluations capability for testing model and RAG performance. You'll learn about Automatic evaluations (Programmatic and LLM as a Judge), Human evaluations with AWS-managed or your own work teams, and RAG evaluations for testing retrieval quality and response generation. We cover key metrics including correctness, completeness, faithfulness, and responsible AI scoring.

In this final console walkthrough, we cover the Settings page including Model permissions and Model invocation logging. We briefly revisit Model access (serverless models are now enabled by default), and point out the User guide and Bedrock Service Terms resources.