Presentation: Finding Your Meme Twin with Embeddings & Vector Databases
summary
Shows how images become 512‑dim embeddings using a Hugging Face image feature extractor in Node.js, stores 72 meme embeddings in Redis as a vector field (flat index or HNSW), and runs local searches on a MacBook without cloud APIs. Demonstrates Euclidean, cosine and inner‑product metrics, a small JavaScript Express client that captures photos, and a working but camera‑fragile demo app with code on GitHub. Highlights semantic caching for LLMs and RAG with vector retrieval.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:53Local AI without cloud costs
- 4:43Understanding embeddings
- 8:00Structured vs unstructured data
- 12:15Image and text embeddings
- 17:33Measuring similarity with vectors
- 22:10Vector database use cases
- 27:48Why not build your own embeddings
- 33:07Neural networks basics
- 40:27Demo setup
- 45:05Live meme twin finder
- 52:15Closing
related talks
52mModern Problems Require Modern Solutions
50mA fun and absurd introduction to Vector Databases
55mVector Similarity Search in Spring with Redis Stack
01h 23mVector Databases Demystified: Discover how they work
46mA Fun & Absurd Introduction to Vector Databases
48mA fun and absurd introduction to Vector Databases
46mA fun and absurd introduction to Vector Databases
01h 01mVector Similarity Search in Spring