Build & deploy generative AI apps in days with Databricks Lakehouse AI
summary
Databricks shows a concrete stack that turns delta tables into production GenAI: vector search syncs any Delta table to a semantic index with automatic updates for millisecond retrieval, feature serving exposes on‑demand structured features as endpoints, and MLflow AI Gateway centralizes model routing and keys so dozens or hundreds of chains can A/B test without changing app code. Block reports retrieval‑based chatbots, inference logging to Delta, and up to ~7x cheaper pretraining via MosaicML while fine‑tuning small models cuts latency and cost.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 0:21Data Centric AI Strategy
- 4:39Enterprise AI Challenges
- 5:17Building AI Applications
- 7:18Vector Search & Features
- 10:39Implementing Vector Search
- 12:33Feature Serving Mechanics
- 13:20Model Leverage Techniques
- 16:23Fine-tuning Embeddings
- 18:58Foundation Model Tuning
- 20:46Pre-training Strategies
- 22:38MLflow AI Gateway
- 25:32Model Serving Unification
- 31:08Lakehouse Monitoring
- 32:08Block's AI Journey
- 32:45Real World Use Cases
- 46:48Summary and Next Steps
related talks
49mGenAI for Business on the Modern Data Stack
23mBuilding AI-Driven Real-Time Lakehouse APIs
07mBuilding and Deploying GenAI Apps at Block
14mTony Baer, dbinsight & Doug Henschen, Constellation Research | Databricks Data + AI Summit 2023
37mOn Building Production-Quality AI Systems
19mDatabricks
26mJoel Minnick, Databricks | Databricks Data + AI Summit 2024
17mHow will AI transform your data platform?