-
Notifications
You must be signed in to change notification settings - Fork 0
Add Real-World Project: RAG Over a GitHub Repo #270
Copy link
Copy link
Open
Labels
area:capstoneCapstone project content, examples, or architectureCapstone project content, examples, or architectureprojectBacklog: a candidate Real-World Project ideaBacklog: a candidate Real-World Project ideatopic:ragRAG, vector search & knowledge graph projectsRAG, vector search & knowledge graph projectstype:content
Description
Metadata
Metadata
Assignees
Labels
area:capstoneCapstone project content, examples, or architectureCapstone project content, examples, or architectureprojectBacklog: a candidate Real-World Project ideaBacklog: a candidate Real-World Project ideatopic:ragRAG, vector search & knowledge graph projectsRAG, vector search & knowledge graph projectstype:content
Add a new Real-World Projects lesson: RAG Over a GitHub Repo (backlog idea #16 — "RAG over your own GitHub repo's code + docs").
Scope
docs/projects/github-repo-rag/index.md— new lesson matching the house skeleton (seedocs/projects/rag-notes/index.mdas the closest sibling in structure: local embeddings, similarity search, free-tier LLM). Builds on the RAG App / rag-notes project.examples/github-repo-rag/— a real, runnable tool: clones (or points at) a local repo folder, chunks code + Markdown docs, embeds them locally withsentence-transformers, and answers questions over the repo with a free-tier LLM, citing the file/line each answer comes from. Includes aquery.py, a sample small repo to clone for testing, anotebook.ipynb, and a README.md.src/data/projects.ts,docs/projects/index.mdx,src/pages/index.tsx(English only — no i18n changes).mainonce merged) — the notebook clones a small real public repo (e.g. this course's own repo or a small sample repo) and runs the pipeline over it, asking for the API key interactively withgetpass.Framing note
The lesson should be honest about chunking code vs. prose (code chunks need line/function boundaries, not just character windows) and about retrieval being approximate: it should encourage spot-checking that cited files actually exist and that the answer is grounded in them.