Purpose of This Repo

This repo is the source of truth for the course project "Customer Support Chatbot with Amazon Bedrock AgentCore" (Prompting for Effective LLM Reasoning). It contains the starter files students use to build the project.

Note: Bedrock Agents Classic was closed to new customers on July 30, 2026. This project runs on its successor, the Amazon Bedrock AgentCore managed harness, with tools exposed through an AgentCore Gateway. Bedrock Evaluations, which the project uses for testing, is unaffected.

Folder Structure

Project Folder

The project folder contains all files and instructions necessary for the project:

• project/README.md — the full project instructions (setup, building the harness, testing, cleanup).
• project/starter/ — the files students start from:
  • CloudFormation templates for the bug-report tool (Lambda + DynamoDB + IAM roles) and the testing resources (S3 + evaluation role)
  • Python setup scripts for the AgentCore resources (setup_gateway.py, create_harness.py), a chat client (chat.py), and cleanup (cleanup_agentcore.py)
  • system_prompt.txt — the student's main deliverable
  • the FAQ document, the evaluation-dataset generator, and a test-suite template

The reference solution, rubric, and detailed docs (docs/tools-setup.md, docs/testing.md) live in the companion solution repo.

What students build

1. Deploy the tool stack (CloudFormation) and create the gateway (setup_gateway.py).
2. Design the system prompt: route each message to bug-report collection, FAQ answering, or a polite human hand-off; collect all bug details across a multi-turn session before filing a ticket with the create_bug_report tool.
3. Create the harness (create_harness.py), iterate with chat.py.
4. Test automatically: run a test suite through generate-eval-dataset.py and score the results with Bedrock Evaluations.
5. Clean up all resources.

All work happens in us-east-1, with the model pinned to us.amazon.nova-pro-v1:0.