Retrieval & RAG

DocPilot

An AI documentation assistant that answers plain-English questions about Hugging Face, PyTorch, and scikit-learn, with every answer traced back to the official docs.

GitHub

Illustration. An open printed document with a paragraph picked out in highlighter, and a small answer card beside it linked back to that paragraph and citing a page number.

What it does

Documentation pages are scraped, chunked, embedded with all-MiniLM-L6-v2, and stored in a local ChromaDB vector store.
A question is embedded the same way, and the top-3 most semantically similar chunks are retrieved.
Those chunks are passed as context to LLaMA 3.1 8B via Groq, which produces a grounded answer returned with its source links.
Streamlit front end over a committed, pre-built vector store.

Built with

Streamlit
ChromaDB
sentence-transformers
Groq / LLaMA 3.1 8B
BeautifulSoup

← All projects