A Retrieval Augmented Generation system to query the scikit-learn documentation
summary
Builds a Retrieval-Augmented-Generation system over scikit-learn docs using an open-source stack (repo: https://github.com/glemaitre/sklearn-ragger-duck), chunking pages with numpydoc and sphinx-gallery, indexing with BM25 lexical and sentence-transformer semantic vectors stored in FAISS, and running LLM inference locally via llama.cpp with a Mistral 7B-style model. Demo answers API and example queries (even with typos) by retrieving disambiguating chunks, but there is no user-data evaluation, no fine-tuning, and hosting/cost for scale remains unresolved.
AI-written summary · about
chapters (11)
- 0:00Introduction and motivation
- 3:41What is RAG
- 7:51Project rules and setup
- 9:58Building the database
- 13:06Handling API documentation
- 15:13User guide and example gallery
- 16:40Lexical vs semantic search
- 18:36Live demonstration
- 20:47Evaluating RAG systems
- 22:55Cost and scalability challenges
- 23:37Q&A session
related talks
50mRetrieval Augmented Generation Beyond Toy Demos
01h 14mAn Introduction to Retrieval Augmented Generation
01h 00mLLMs gone wild
43mAdvanced Retrieval-Augmented Generation Techniques
51mDeveloping a Custom RAG System
02h 40mFrom RAGs to riches: Build an AI document inquiry web-app
25mA Deep Dive Into Retrieval-Augmented Generation with LlamaIndex
28mImproving responses in generative AI applications