Self-Service Analytics using LLMs
summary
A concrete recipe for a production-ready LLM-based self-serve analytics platform: connect a data catalog connector, use an LLM as query generator to emit SQL, run results through a python code generator for visuals, and guard outputs with ACLs and query validation. Notes on what works and fails: RAG helps at scale, latency and non-determinism persist, accuracy correlates with documentation quality, smart retries and context injection improve results, and prompt injection requires explicit mitigation. Example uses a Chinook-style toy DB to show end-to-end flow.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:26Self-service analytics
- 3:55Self-service challenges
- 7:51Large language models
- 11:22LLMs for analytics
- 13:10LLM analytics architecture
- 20:27Example walkthrough
- 24:00LLM limitations
- 27:35Improving LLM analytics
- 30:04Security considerations
- 31:49Resources
- 32:24Q&A
related talks
38mLLMs for Self-serve Analytics - Lessons Learned
40mLLMs & Data Warehouses: Reliable AI Agents for Business (Keynote)
15mFrom Words to Data: Using LLMs to Turn Questions into SQL
24mDemocratizing data access through LLM
47mGive Your LLMs a Left Brain
47mA Step-by-Step Guide to Securing Large Language Models (LLMs)
52mAnatomy of an LLM
22mPractical Strategies for Implementing Domain-Specific LLMs