Retrieval Augmented Generation
summary
Retrieval Augmented Generation combines embeddings and vector search to cut hallucinations by feeding an LLM retrieved, up‑to‑date snippets; the demo connects three sources (unstructured destination texts, an SQL table of carnival visitor counts, and a realtime weather API) and autonomously chains function calls to fetch largest‑carnival info and current weather for Rio de Janeiro, producing correct answers. Implementation uses LlamaIndex, stores vectors with pgvector on Postgres in the project (they note Qdrant as highest‑performance), measures similarity with cosine, and uses PromptFu for end‑to‑end tests.
AI-written summary · about
chapters (14)
- 0:00Introduction to Retrieval Augmented Generation
- 1:10Motivation and challenges
- 3:17RAG vs hallucinations and outdated knowledge
- 5:23Building knowledge base for semantic search
- 7:12Semantic vs keyword search
- 8:15Encoding documents for vector search
- 10:01Vector databases in RAG
- 11:46Searching knowledge base with queries
- 13:13Extending RAG with AI agents
- 15:40AI agents orchestrating data sources
- 16:26Demo setup and data sources
- 18:37Live RAG system demo
- 22:54Conclusion and takeaways
- 23:18Q&A
related talks
50mRetrieval Augmented Generation Beyond Toy Demos
01h 14mAn Introduction to Retrieval Augmented Generation
33mVector Databases for GenAI applications
22mFrom Postgres full text search to Retrieval Augmented Generative search
51mDeveloping a Custom RAG System
49mGenerating the Future
48mAn Introduction to Retrieval Augmented Generation
33mVector Databases in the Age of AI: Embracing RAG for Enhanced Search Capabilities