Vector Databases Demystified: Discover how they work
summary
Shows concrete results not fluff: a runnable repo (create-a-vector-db) that builds an in-memory vector store using all-miniLM L6 v2 384‑dim embeddings, a brute‑force top‑5 cosine search over ~50 items, and a RAG demo that replaces hallucinations by passing retrieved product text into an LLM. Compares cosine vs Euclidean, context windows (4K→8K→1M+), hosting costs ($900/mo GPU vs cents on managed Gemini), and a scalable production path using HNSW hierarchical navigable small world indexes shipped in Aerospike.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:08LLMs and prompt engineering
- 10:24RAG applications and hallucinations
- 26:05Vectors and semantic search
- 37:40Build a vector DB from scratch
- 51:00Vectors with LLMs for answers
- 1:06:15Brute-force search limits
- 1:11:32HNSW indexes for scalability
- 1:18:03Dedicated vs multimodal systems
- 1:21:16Real-world applications
- 1:22:44Q&A
related talks
39mWhat's in a Vector Database?
50mA fun and absurd introduction to Vector Databases
41mGetting Started With Vector Databases
48mA Fun & Absurd Introduction to Vector Databases
46mA fun and absurd introduction to Vector Databases
35mImpact of Vector Search
48mA fun and absurd introduction to Vector Databases
01h 46mVector Databases (for GenAI & beyond)