Developer documentation assistant with sources

The brief

An open-source project gets the same questions again and again in its forums, though the answers are in the documentation. Build an assistant over the project's docs that answers with links to the exact pages, refuses when the docs do not cover a question, and is proven with an evaluation, not a demo. Good looks like: a document pipeline, two retrieval setups compared on at least 50 test questions, citations on every answer, and a deployed app.

What you will build

1. Collect and clean the documentation; chunk it with page and section metadata
2. Build a baseline vector search and a hybrid search with re-ranking
3. Write 50+ test questions with expected sources
4. Evaluate faithfulness, answer relevance and retrieval recall for both setups
5. Add "I don't know" handling and citations to the exact page
6. Deploy the assistant and write up the trade-offs

Free, open data to build it with

• FastAPI documentation source ↗ FastAPI (GitHub)
• Python documentation downloads ↗ Python Software Foundation

Check each dataset's licence before you publish your work. Your mentor gives your team its own version of the brief (data, scope or region), so every team's project is its own.