Generative AI on AWS training: Amazon Bedrock, RAG and agents

What you'll build

By the end of day two you have a working AI assistant app: a frontend, a serverless backend on AWS Lambda and Amazon Bedrock as the model. It answers questions from your own documents (Knowledge Bases and RAG), has Guardrails configured and is deployed with AWS SAM, so you can recreate it in any account.

Part of the backend from lab 3 (Python, AWS Lambda, Converse API):

import boto3, json, os

bedrock = boto3.client("bedrock-runtime", region_name="eu-central-1")
MODEL_ID = "eu.anthropic.claude-opus-5-5"   # Claude Opus 5.5, EU inference profile
GUARDRAIL_ID = os.environ["GUARDRAIL_ID"]

def lambda_handler(event, context):
    prompt = json.loads(event["body"])["prompt"]
    response = bedrock.converse(
        modelId=MODEL_ID,
        messages=[{"role": "user", "content": [{"text": prompt}]}],
        inferenceConfig={"maxTokens": 1000},
        guardrailConfig={"guardrailIdentifier": GUARDRAIL_ID, "guardrailVersion": "1"},
    )
    answer = response["output"]["message"]["content"][0]["text"]
    return {"statusCode": 200, "body": json.dumps({"answer": answer})}

You get the code from every lab after the course.

Agenda

1. Introduction

   AWS basics and application security recap

   Machine learning basics: just enough to understand what a model does

   What generative AI is and where it fits in a product

   Lab · Setting up your AWS environment for GenAI development
2. How GenAI apps are built

   From idea to working app: methods and approach

   Components of a GenAI app and choosing a model

3. Amazon Bedrock

   Models in Bedrock: Anthropic, Amazon, Meta and others through one API

   Working with LLMs and fitting a model to the task

   Common use cases, costs, security and data privacy

   Lab · Working with LLMs in the AWS console
4. Running the app on AWS

   Compute options for a GenAI backend

   AWS Lambda: development, testing, Lambda Durable Functions

   Deploying with AWS SAM

5. Prompt engineering

   What a prompt is and how to build one

   Techniques: roles, examples, output format, chaining

   Prompt misuse and bias mitigation

   Lab · A simple Q&A app with Amazon Bedrock and Guardrails
6. Architecture

   Example architectures: chatbot, Q&A over documents, agent

   Components: model, RAG, Knowledge Bases, Guardrails, agents

   Good design practice and the mistakes that cost money

7. Tools and frameworks

   LangChain

   Strands Agents SDK

   Model Context Protocol (MCP)

   Lab · An app with Knowledge Bases and RAG, built with LangChain, protected by Guardrails and deployed with AWS SAM
8. Wrap-up

   Key takeaways and next steps: what to add to your app after the course

Programme last updated: 26 September 2026

Price and package

Why train the whole team

One group, one language. After two days everyone on the team understands the same terms and has seen the same code, so the conversation about AI in your product starts with architecture, not definitions. Instead of a year of planning you have a working prototype you can put numbers and risks against.

Your trainer

Trainer

Łukasz Dorosz

AWS Community HeroAWS Champion Authorized InstructorCloud Solutions Architect

An architect who teaches from real projects. He works on serverless architecture, AWS security and generative AI, and helps developers and architects get confident with AWS and ship AI apps that hold up in production.

Co-organiser of AWS Community Day Poland and AWS User Group Warsaw, and creator of the CloudHeroes YouTube channel.

15+

years in IT

8+

years in AWS

8

AWS certifications

More: about me · AWS architecture and consulting

How the training works

Common questions

1. Do I need to know AWS to take part?

   The basics, yes: the console, IAM, running a Lambda function. Module 1 recaps the key services, but the course assumes your team has worked with AWS before. If not, start with AWS Introduction or AWS for developers and GenAI, which ends with your first chatbot on Bedrock.

2. Which tools do we use?

   Amazon Bedrock, Knowledge Bases, Guardrails, AWS Lambda with Durable Functions, AWS SAM, Python, LangChain, Strands Agents SDK and the Model Context Protocol (MCP). You get all the lab code after the course.

3. Online or on site?

   Either, with the same programme and the same labs. Online sessions are live, usually two days of 8 hours. On site, I come to your office.

4. How is Amazon Bedrock different from calling OpenAI or another API directly?

   Bedrock is a managed AWS service that gives you models from several providers (Anthropic, Amazon, Meta and others) through one API. Your data stays in your AWS account, and you get the IAM, CloudTrail, VPC and compliance controls you already use for the rest of your infrastructure, plus Guardrails, Knowledge Bases and model customisation. On the course I show when that matters and when it doesn't.

5. How much does it cost to run a GenAI app on AWS?

   Bedrock models are billed per input and output token, so the cost depends on traffic and prompt length. An AI assistant with moderate traffic usually costs a few tens of dollars a month. On the course we go through the ways to bring that bill down: shorter prompts, response caching, picking the right model for the job and cost monitoring.

6. What is RAG, and why do I need a Knowledge Base?

   RAG (Retrieval Augmented Generation) lets a model answer from your own documents, databases and APIs without training it. A Bedrock Knowledge Base handles the indexing and search for you. It's the standard way to make a business app answer questions about current data, not whatever the model remembers from a year ago.

7. Can the programme be adapted to our team?

   Yes. Before the course we talk about what your team already knows and what you want to build. I usually rebalance the modules or swap a lab example for a case from your own product.

Next step

Need solid serverless basics on AWS first? Three days for developers, including your first chatbot on Bedrock.

Other AWS courses: AWS Introduction, AWS Specialist. All courses: AWS training.