GenAI Beyond Prototyping: The Path to Production with AI-Native Databases
summary
Demo shows an AI‑native database called W8 powering a Streamlit RAG prototype that runs on a local model or API and returns searches in ~300 ms and claims a billion‑object filtered search at ~600 ms. It uses an HNSW index (in‑memory, fast but memory heavy), applies product quantization and binary quantization to cut vector RAM, reporting ~6 GB → 2 GB for 1M vectors, cites 10M ≈ 65 GB RAM, covers sharding, replication, multi‑tenancy, backups and cost math ($5/hr → $440k/yr).
AI-written summary · about
chapters (12)
- 0:00GenAI production challenges
- 3:20GenAI prototype demo
- 7:42RAG limitations recap
- 10:11Infrastructure for GenAI
- 12:00Scaling with HNSW indexes
- 15:35Memory in vector DBs
- 21:36Vector compression
- 24:27Horizontal scaling strategies
- 25:51Multi-tenancy
- 29:03Reliability and fault tolerance
- 32:52Transition to production
- 35:23Q&A: scaling and latency
related talks
13mRAG Without the Hassle: Building AI-Powered Applications
01h 35mTutorial: Build AI-Supercharged RAG Apps with a Vector Database
19mPOC to PROD: Hard Lessons from 200+ Enterprise GenAI Deployments
24mScalable GenAI in Production
47mRAG at scale: production-ready GenAI apps with Azure AI Search
56mDeveloping AI Applications with SQL AI: Ground to Cloud to Fabric
53mModern PostgreSQL for AI...
20mOpen Source Database Operations in the Age of AI