Pippy Debugger backend functionality

[iris-wuu]

I created three functions in this PR:

1. prompt_template() integrated #102 prompt engineering into backend script.
2. code_analysis_with_llm() maintains chat history by appending new requests / responses and then send chat history to LLM to get contextual responses.
3. get_ai_response() calls the 2 functions above. This should be called in frontend script as a connection between frontend and backend.

Reference: https://platform.openai.com/docs/guides/text?api-mode=responses#page-top

[iris-wuu]

Hi project mentors! @walterbender @chimosky @quozl

I wrote about the backend's high-level logic in my proposal but didn't include detailed implementation. I may submit more PRs in the coming days. Hope these could also be taken into account when evaluating my proposal.

I didn't get a chance to discuss the llm selection when writing my proposal. However, I recognized it's quite necessary as I saw some AI agents in sugar labs were implemented with open-source local LLM and FastAPI.
Do we prefer local LLM rather than GPT4?
Do we prefer a full RAG pipeline or using LLM alone to generate responses?

If it is the case, I can try modifying my implementation.
Would love to hear about your opinions!

[quozl]

Thanks. I'm not a mentor this year. I've not looked at your proposal, sorry. My opinion is that Sugar activities should work without internet where it is practical to do so.

[chimosky]

In addition to what Quozl has said, they're other ways to contribute to our software to show us your skills, you can start with existing issues.

[iris-wuu]

Thank you both for the comments!