Leveraging Cassandra to Provide Generative AI That Powers Enterprise Analytics
summary
Skypoint builds an enterprise AI lakehouse that combines Databricks structured tables, vector retrieval over Astra DB, LangChain agents and fine‑tuned LLMs to answer long‑form QA with sub‑5‑second SLAs; production since end of June and early customers report 85–95% accuracy and large nurse time savings that map to the uploader's 30–50% productivity target. Techniques include self-reflection SQL generation, multi‑query decomposition and parallelized RAG over PDFs and tables; hallucination mitigation and observability are implemented and latency tradeoffs drove the move to Cassandra/Astra.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:07Elderly care crisis impact
- 3:15Skypoint AI platform overview
- 6:06SQL generation with React
- 8:36Handling structured queries
- 10:21Fine-tuning and latency optimization
- 11:48Combining data retrieval methods
- 13:15Production deployment benefits
- 15:06Hallucinations and SLA challenges
- 17:15Q&A
related talks
18mKeynote: Data = AI: Leading the Future of Gen AI with Apache Cassandra
23mTrusted Agentic AI:Architecting Systems with Reliable Data at Scale
17mAI in Production: Lessons Learned
18mAccelerate generative AI development with integrated data platforms
24mScalable GenAI in Production
40mInsights and Key Learnings on Our AI Journey
45mForget SQL! Create your own AI Data Analyst
19mHow to use generative AI to deliver self-service data products