Scaling personalization with AI
summary
Demo builds a vanilla RAG pipeline over GitHub issues using Hugging Face models, LangChain orchestration, a FAISS vector DB and a quantized Mistral 7B LLM. Documents are chunked into 512-token chunks with 30-token overlap, embedded, and nearest-neighbor retrieval (k=4) is stuffed into prompts. Claim: combining retrieval plus a good model cuts hallucinations from ~45% to ~5%. The demo runs but gives inconsistent answers when source data is noisy, so input quality and retriever tuning—not model choice alone—turn out to be the main limit.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 3:38AI and ML overview
- 12:54Hugging Face and vector DBs
- 20:50Presenters and demo setup
- 27:18Retrieval augmented generation architecture
- 31:12Vanilla RAG workflow demo
- 44:25Demo results and analysis
- 54:30Improving RAG systems
- 1:02:46Deployment and Q&A
related talks
41mFrom RAGs to riches: Build an AI document interrogation app in 30 mins
25mSet Up an Open Source RAG in 30 Minutes
43mLarge Language Models for Devs: from zero to your first LLM application
27mOptimizing RAG Systems in Practice: Challenges and Proven Solutions
28mNavigating Challenges and Enhancing Performance of LLM based Applications
35mThe Hidden Jewels to Solve Confabulation Challenges in Generative AI’s RAG Promise
01h 33mRAG How To Talk To Your Data
27mFrom RAGs to riches: Build an AI document interrogation app in 30 mins