Report Generation: Switching from OpenAI SDK to Google ADK

[lotif]

Summary

Switching from OpenAI SDK to Google ADK on Report Generation agent to match the other agents.

Some extra code was needed to parse the response of the agents which is a bit trickier with Google ADK.

Clickup Ticket(s): NA

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

• Switching from OpenAI SDK to Google ADK on report generation agents
• Modifying the demo and evaluate code accordingly
• Added an event parser to parse the response of the google models
• Removing OpenAI dependencies from pyproject.toml
• Updating .env.example and the Config class to remove OpenAI env vars

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 demo UI and the evaluation script

Checklist

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

[fcogidi]

As a point of discussion: I think we should keep this and use it for single LLM calls (e.g. llm-as-a-judge evaluation). The reason is that the openai client is almost standard and can work with multiple providers, including gemini, by changing the base url and the api key. I don't think the client provided in the google genai library has that level of compatibility.

I'm working on a reusable llm judge implementation that uses this client. So, the change doesn't need to be made in this PR.

@amrit110 Please chime in here as well

[amrit110]

Yeah that makes sense. We are just using the Google genai client because we are using gcp for the bootcamp, but I was actually telling Ethan we may use this repo for some applied research work as well, where we might switch out the judge model. So good to keep using the openai client.

[lotif]

Ok that makes sense. I also prefer the OpenAI API, the calls and the returns are way simpler to work with and the trace reports to langfuse were better.

I'll revert this change here and merge since you will need it later.

[amrit110]

Seems good.