Databases in the AI Trenches
summary
Describes embedding workflow: chunk text, fetch OpenAI embeddings, average token vectors, store them in Postgres (example uses pgvector), then run nearest‑neighbor searches to return the closest sentence — e.g., “who is the king” matches “the king is a tall man,” while “who is pretty” matches “queen” because embeddings relate words. Shows RAG by pulling JSON rows from the database, prefixing them to prompts so the model answers specific invoice questions and even generates SQL. Notes training and transformers are poor fits for databases.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:52Outline
- 3:20AI explosion
- 6:43Hyper-dimensional vectors
- 8:14Training text embedding
- 11:46Semantic/vector searc
- 16:21Transformers
- 17:55Generative AI
- 18:42Retrieval-augmented generation (RAG)
- 21:18Deployment options
- 23:03Conclusion
related talks
01h 46mVector Databases (for GenAI & beyond)
06mWhat makes vector databases essential for RAG and AI apps?
27mPutting AI in Production
48mUnderstanding Embedding Vectors and the RAG Pattern in AI Solutions
44mIntegrating AI with Postgres...
36mUnleashing the Power of Vector Databases for AI-Driven Applications
12mVector Databases: Why QA professionals Needs to Care About them in the Age of AI?
45mUsing Postgres + OpenAI to power your AI Recommendation...