How to Decipher User Uncertainty with GenAI and Vector Search
summary
Shows how BERT and GPT embeddings plus vector indexes turn vague queries into semantic matches, not keyword hits, and compares distance metrics (Euclidean, cosine, Manhattan, dot product) by use case. Describes a concrete build: a serverless function returning 5 of 35 blog posts in ~1–2s using a vector index hosted on a free 10 GB tier and RAG to reduce hallucination. Notes a production example: Sherlock: 50 ms fraud detection for 12M users.
AI-written summary · about
chapters (14)
- 0:00Introduction and setup
- 2:14Problem with traditional search
- 5:06What are vector embeddings
- 8:14Neural networks and embeddings
- 11:30Vector search and closeness
- 13:18Measuring vector similarity
- 17:10When to use vector search
- 20:04Chatbots and hallucinations
- 21:31Retrieval augmented generation
- 22:34Use cases for vector search
- 23:42Real-time fraud detection
- 25:10Example with personal blog
- 28:20Building your own demo
- 29:02Book announcement and closing
related talks
57mEnter the Brave New World of GenAI with Vector Search
48mUnderstanding Embedding Vectors and the RAG Pattern in AI Solutions
28mUse OpenSearch as Your Vector Search Database To Optimize GenAI Accuracy!
19mPOC to PROD: Hard Lessons from 200+ Enterprise GenAI Deployments
18m: Workforce Efficiency with AI-Driven Vector Search
55mSupercharged Search with Semantic Search and Vector Embeddings
45mTomorrow’s SEO: How AI is Reshaping Search
39mUsing Dense Vector search at the EU Publications Office