Cloud · CLC-09

Generative AI Apps and Agents on Amazon Bedrock

Generative AI Apps and Agents on Amazon Bedrock is a hands-on developer course in building generative AI applications on AWS with Python, from the Converse API, RAG with Bedrock Knowledge Bases and Guardrails to AI agents on AgentCore. It suits developers and cloud engineers taking AI into production, and you leave with a working internal AI assistant prototype.

Updated 1 Oct 2026

From 8,910 THB / person 9,900 −10% excl. VAT 7% · group rates available

Corporate training quote Join the waiting list

PDFDownload the course outline

Duration18 hours · 3 days
FormatOnsite / live online
Next roundOn request
CertificateIncluded

Course overview

Building a generative AI chatbot prototype takes a few hours, but taking the same application into production in an organisation is another matter. Development teams have to decide which model to use, what each request costs, whether answers cite the right documents, how to stop personal data leaks and prompt injection, and, once AI agents can call other systems on their own, how to control their permissions and trace what they do. Amazon Bedrock offers services for all of this, but you need to know how to fit the pieces together.

This course takes developers through building generative AI applications on AWS with Python from start to finish. It begins with calling models through the Converse API, streaming, tool use and cost control, moves on to RAG with Amazon Bedrock Knowledge Bases, rerankers and quality measurement, protects the application with Amazon Bedrock Guardrails, IAM and logging, and then builds AI agents with Bedrock Agents and Strands Agents before deploying them to Amazon Bedrock AgentCore with Memory, Gateway, Identity and Observability. It closes with a capstone internal AI assistant. The content helps build skills aligned with the AWS Certified Generative AI Developer - Professional (AIP-C01) exam. (3 days, 6 hours per day, 18 hours in total, Intermediate level.)

What you’ll gain

Call foundation models on Amazon Bedrock with the Converse API, including streaming, multimodal input and tool use
Choose models and design calls cost-effectively with prompt caching, batch inference and cross-Region inference
Version prompts and test their quality before releasing them
Build RAG with Bedrock Knowledge Bases and improve it with chunking, metadata and rerankers
Guard against harmful content, personal data exposure and hallucination with Bedrock Guardrails
Build AI agents with Bedrock Agents and Strands Agents and connect tools through MCP
Deploy agents to AgentCore Runtime with Memory, Gateway, Identity and Policy
Monitor, evaluate and control the cost of generative AI applications in production

Who this course is for

Backend and full-stack developers building generative AI applications on AWS
Cloud engineers and solutions architects designing RAG systems or AI agents for their organisation
AI and innovation teams that need to take prototypes into production safely
Anyone building skills aligned with the AWS Certified Generative AI Developer - Professional exam
People who have passed AWS AI Practitioner and want to build real applications

Prerequisites

Python programming, including functions, dictionaries, JSON and installing packages
Basic use of the AWS console and a working knowledge of IAM, S3 and Lambda
A basic grasp of LLMs and prompts; having passed AWS AI Practitioner makes the course smoother
A laptop that can run Python 3, VS Code and the AWS CLI, plus an AWS account on a Paid Plan

Curriculum

Course Details

This is a developer course in building generative AI applications and AI agents on Amazon Bedrock. It runs for 3 days, 6 hours per day (18 hours in total, 09:00-16:00), as lectures with Python coding labs every day. Intermediate level. Every lab runs in the learner's own AWS account or a company sandbox on a Paid Plan, with usage charges on that account; the course uses small models, S3 Vectors by default and clean-up

steps after each lab. It builds on the RAG & Knowledge Base course and focuses on AWS managed services from RAG through to agents in production. The content is aligned with some of the skills in the AWS Certified Generative AI Developer - Professional (AIP-C01) exam, but it is not official AWS exam preparation and does not include the exam fee. Learners take home a lab guide, code for every lab and a capstone internal AI assistant project.

Day 1 Calling Foundation Models Professionally

Section 1: Lab: Preparing Your AWS Account and Amazon Bedrock

Generative AI services on AWS: how Bedrock, AgentCore and SageMaker AI differ
Choose a Region, check model access and understand cross-Region inference profiles
Create least-privilege IAM roles and policies for developers and applications
Set up the AWS CLI, boto3 and local credentials safely
Lab: set AWS Budgets and call a model from Python for the first time

Section 2: Lab: Calling Models with the Converse API

Converse API structure: system prompt, messages and inference configuration
Send images and documents to a model as multimodal input
Stream responses with ConverseStream for a smoother user experience
Handle throttling and errors with exponential backoff retries
Lab: build a summary API that accepts both text and PDF files

Section 3: Choosing Models and Controlling Cost

Compare Bedrock model families on quality, speed, language and price
Calculate cost per request from input and output tokens
Cut costs with prompt caching, batch inference and smaller models
On-demand versus provisioned throughput, and designing model routing
Lab: measure quality, latency and cost of three models on the same task

Section 4: Lab: Prompt Engineering and Prompt Management

Write system prompts and few-shot examples and enforce JSON output
Store prompts as templates with variables and versions in Bedrock Prompt Management
Chain several steps into a workflow with Amazon Bedrock Flows
Run prompt regression tests before changing a version in production
Lab: version the prompts of a customer request classification task

Section 5: Lab: Tool Use and Function Calling

Define tools with JSON Schema and let the model decide when to call them
Write the loop that receives tool requests, runs functions and returns results
Validate parameters and handle tool errors so the app stays up
Use AWS Lambda as a tool that connects to internal systems
Lab: an assistant that answers order status questions from a mock system

Day 2 RAG, Guardrails and Security

Section 6: Lab: Embeddings and Vector Stores on AWS

How embeddings work, and choosing an embedding model and dimension
Vector store options: S3 Vectors, OpenSearch Serverless and Aurora PostgreSQL with pgvector
Compare minimum cost, speed and capabilities of each option
Similarity search, metadata filters and hybrid search
Lab: embed Thai documents and run similarity search in Python

Section 7: Lab: RAG with Amazon Bedrock Knowledge Bases

Create a knowledge base from Amazon S3 and choose a vector store
Fixed-size, hierarchical and semantic chunking and how to choose
Parse documents with tables and images using a foundation model or Bedrock Data Automation
Retrieve versus RetrieveAndGenerate, with citations in the answer
Lab: a chatbot that answers from the company handbook with citations

Section 8: Improving and Measuring RAG Quality

Add metadata to documents to filter by department, date or permission
Improve accuracy with reranker models and query reformulation
Sync new content incrementally when source documents change
Measure RAG quality with Amazon Bedrock Evaluations and LLM-as-a-judge
Lab: compare results before and after tuning chunking and reranking with a test question set

Section 9: Lab: Amazon Bedrock Guardrails

Content filters, denied topics and word filters for business applications
Sensitive information filters to mask or block personal data
Contextual grounding checks and Automated Reasoning checks to reduce hallucination
Use the ApplyGuardrail API with any model and defend against prompt injection
Lab: add a guardrail to the RAG chatbot and test it with attack prompts

Section 10: Security and Governance for Generative AI Applications

Least-privilege IAM for models, knowledge bases and agents
Encryption with AWS KMS and private connectivity through VPC endpoints
Model invocation logging, CloudTrail and keeping logs without exposing personal data
Using company data with models in line with PDPA and internal policy
Workshop: a security checklist before opening a generative AI app to real users

Day 3 AI Agents from Prototype to Production

Section 11: Lab: AI Agent Concepts and Bedrock Agents

How agents differ from chatbots: reasoning, tools, memory and planning
ReAct, fixed-step workflows and multi-agent patterns
Managed Bedrock Agents with action groups and knowledge bases
Choosing between managed agents and building your own with a framework
Lab: build a Bedrock agent that searches the handbook and opens tickets through Lambda

Section 12: Lab: Building Agents with Strands Agents

Strands Agents SDK structure: model, system prompt and tools
Create tools with a Python decorator and use tools from an MCP server
Multi-agent patterns: agents as tools, graph and swarm
Set stop conditions, timeouts and permission boundaries for agents
Lab: an agent that analyses sales data by calling several tools

Section 13: Lab: Deploying with Amazon Bedrock AgentCore

AgentCore overview: Runtime, Memory, Gateway, Identity, Policy and Observability
Deploy an agent to AgentCore Runtime with the AgentCore CLI
Add short-term and long-term memory so the agent remembers user context
Turn Lambda functions and internal APIs into MCP tools through AgentCore Gateway
Authenticate and restrict tool calls with AgentCore Identity and Policy

Section 14: Observability, Evaluation and Operations

Trace each step of an agent with AgentCore Observability and CloudWatch
Measure agent quality with AgentCore Evaluations and the team's own test sets
Set alarms for errors, latency and unusual token usage
CI/CD and rollback approaches when changing models or prompts
Lab: find why an agent answered wrongly from its trace and fix it

Section 15: Capstone: An Internal AI Assistant

Design an architecture that combines RAG, guardrails and an agent
Build an assistant that answers from documents and acts through tools
Evaluate quality, safety and cost per use
Present the work and map the skills used to the AIP-C01 exam domains
Delete all resources and wrap up with a pre-production checklist

Schedule & training options

For individuals — public rounds

No public rounds are open right now. Join the waiting list and we will contact you first when the next round opens, or ask us on LINE. Or call 02-570-8449 or 088-807-9770

Join the waiting list Ask about rounds

For organisations — in-house / private

Tailor the content to your team’s tools and projects
Your dates, at your office or live online
Quotation with tax ID for procurement

Corporate training quote

Instructors

Samit KoyomFounder & CEO, IT Genius Engineering

Frequently asked questions

What is the Amazon Bedrock Converse API, and how does it differ from calling models the older way?

The Converse API provides one consistent way to call any supported model on Amazon Bedrock, so you can switch models with minimal code changes, unlike the older approach of formatting requests to each provider's own specification. It supports multi-turn conversations, system prompts, image and document inputs, streaming responses and tool use, which lets the model call your own functions.

What are the steps for building RAG with Bedrock Knowledge Bases?

Store your documents in S3, create a knowledge base, choose an embedding model and a vector store such as S3 Vectors, and set a chunking strategy suited to your documents. Run a sync to convert the documents into vectors, then call it through the API to retrieve content and have the model answer with source citations. Improve quality further with metadata filters and a reranker.

How can you reduce the cost of a generative AI app on Amazon Bedrock?

Choose the smallest model that does the job well enough rather than a large model for everything. Use prompt caching for parts of the prompt that are sent repeatedly, such as system prompts or long documents, use batch inference for work that does not need immediate results, keep input and output lengths in check, and monitor token usage regularly to find waste.

Who is the Generative AI Apps and Agents on Amazon Bedrock course for, and what should I know beforehand?

It suits backend and full-stack developers building generative AI apps on AWS, cloud engineers and solutions architects designing RAG or agent systems, AI teams taking prototypes into production, and people who have completed AWS AI Practitioner. It is intermediate level: you should write Python, use the AWS Console, understand IAM, S3 and Lambda at a basic level, and grasp LLM and prompt concepts.

What will I be able to do after the Generative AI Apps and Agents on Amazon Bedrock course?

You will be able to call models with the Converse API using streaming and tool use, control costs with prompt caching and batch inference, build RAG with Bedrock Knowledge Bases, guard against personal data exposure and hallucinations with Guardrails, build agents with Strands Agents connected to tools through MCP, and deploy them to AgentCore Runtime, finishing with an internal AI assistant project. Participants receive a Certificate of Completion with its own number that can be verified online.

What do I need to prepare for the Generative AI Apps and Agents on Amazon Bedrock course?

Bring a laptop on which you can install Python 3, VS Code and the AWS CLI, plus your own paid-plan AWS account or an organisational sandbox. Usage is charged to your account at actual consumption; the course uses small models, defaults to S3 Vectors and includes steps to delete resources after each lab. The content aligns with some skills in the AIP-C01 exam but is not official AWS exam preparation.