How to Construct and Assess Your First LLM Based RAG System
summary
Builds a working RAG chatbot that scrapes the Hugging Face NLP course into 95 documents, chunks with a recursive splitter, embeds with MiniLM L6 v2 embeddings, indexes via a FAISS-style vector store and retrieves top‑2 chunks, then generates with Llama 3.1 instruct (8B) hosted locally on a GPU (A100). System returns "I don't know" when retrieved context lacks an answer. Evaluation uses Trulens/RAGas with a judge LLM; observed metrics: answer relevance ≈0.07, context recall ≈0.5–0.75, groundness ≈0.28.
AI-written summary · about
chapters (13)
- 0:00Workshop introduction
- 2:01Language model basics
- 9:38RAG for private data
- 13:30What is RAG
- 18:56Setup and model download
- 27:12Scraping course data
- 34:18Chunking documents
- 46:02Embeddings and vector DB
- 51:21RAG prompt template
- 1:07:11RAG chain in LangChain
- 1:12:24Evaluating with TruLens
- 1:23:49Improving RAG
- 1:30:07Summary and resources
related talks
34mEvaluating LLM Frameworks
18mLLMs and GenAI: A Guide to Building Your Own RAG Model for Document Interaction
31mBeyond LLMs and Maturing RAG use cases
44mWhat Is a RAG System and How to Build One
38mEmpowering AI Education with Rag: Building Domain-Specific LLM Agents for Enhanced...
48mCrafting your own RAG system: Leveraging 30+ LLMs for enhanced performance
42mTake Your RAG To The Max!
40mPractical Lessons from Deploying AI Assistants Based on LLMs