Add notebooks for AML investigation use case

[fcogidi]

Summary

Add notebooks for AML investigation use case.

Clickup Ticket(s): N/A

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 💥 Breaking change (fix or feature that would cause existing functionality to not work as expected)
• 📝 Documentation update
• 🔧 Refactoring (no functional changes)
• ⚡ Performance improvement
• 🧪 Test improvements
• 🔒 Security fix

Changes Made

• Add notebook to explore AML dataset and database tools.
• Add notebook to explain how cases are built and how to run the agent on a case file.
• Add notebook to showcase evaluation pipeline and explain metrics.

Testing

• Tests pass locally (uv run pytest tests/)
• Type checking passes (uv run mypy <src_dir>)
• Linting passes (uv run ruff check src_dir/)
• Manual testing performed (describe below)

Manual testing details:
Ran the notebooks.

Screenshots/Recordings

N/A

Related Issues

N/A

Deployment Notes

N/A

Checklist

• Code follows the project's style guidelines
• Self-review of code completed
• Documentation updated (if applicable)
• No sensitive information (API keys, credentials) exposed

[Copilot]

Pull request overview

This PR adds three comprehensive Jupyter notebooks that document and demonstrate the AML (Anti-Money Laundering) investigation use case for evaluating AI agents. The notebooks provide a step-by-step walkthrough from data exploration to running evaluations.

Changes:

• Added notebook 01 to explore the IBM AML dataset, build a SQLite database, and demonstrate the ReadOnlySqlDatabase tool
• Added notebook 02 to explain case file structures, case generation, and running the agent on individual cases
• Added notebook 03 to showcase the full evaluation pipeline with item-level, trace-level, and run-level graders

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

File	Description
implementations/aml_investigation/01_data_and_tools.ipynb	Introduces the AML dataset, demonstrates database setup and schema, and explains the ReadOnlySqlDatabase safety tool
implementations/aml_investigation/02_running_the_agent.ipynb	Documents the case file data structures, explains the four case types (TP/TN/FP/FN), and demonstrates running a single case through the agent
implementations/aml_investigation/03_evaluation.ipynb	Demonstrates the full evaluation pipeline including dataset upload to Langfuse, explains the three-tier grading system, and shows how to inspect results

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[amrit110]

Great work @fcogidi! Can you actually try running everything from your implementations once on a coder workspace? I think if it all works, then merge this PR, and we should be good :)

[fcogidi]

Great work @fcogidi! Can you actually try running everything from your implementations once on a coder workspace? I think if it all works, then merge this PR, and we should be good :)

It works in Coder. I just had to change the OPENAI_API_KEY envvar to GOOGLE_API_KEY for the agent calls to work. I think this is an environment variable that is not explicitly passed in, so it's looking for the canonical names internally.

I can do this as a workaround: os.environ["GOOGLE_API_KEY"] = Configs().openai_api_key.get_secret_value()

I'd recommend changing to GOOGLE_API_KEY in the .env file and updating firestore.