From Words to Data: Using LLMs to Turn Questions into SQL
summary
Claims that LLMs convert natural language to SQL in seconds, replacing hours or days, and demonstrates this end-to-end with a Jaffle Shop demo that actually runs generated queries. Uses dbt semantic layer to supply table/metric definitions, tests with ChatGPT and shows hallucination drops, scales via retrieval-augmented generation to avoid huge prompts, and orchestrates retriever/generator/executor/narrator agents. Presenter cites a claimed 15% change in returning-user metric as an example of surprising business impact.
AI-written summary · about
chapters (8)
- 0:00AI-powered data questions
- 2:38LLM limitations without data
- 5:30Semantic layers with DBT
- 8:23SQL with semantic context
- 10:07Scaling with retrieval augmented generation
- 11:57SQL generation system
- 12:40Benefits for teams
- 13:24Future of conversational analytics
related talks
40mLLMs & Data Warehouses: Reliable AI Agents for Business (Keynote)
24mDemocratizing data access through LLM
38mLLMs for Self-serve Analytics - Lessons Learned
24mHarnessing the Power of LLMs in Advanced Analytics
35mFunction Calling for LLMs
41mEnhancing LLMs with Graph Technology
30mThe Future of Data is Words
33mOpening Keynote: LLMs, between fantasy and reality - the poin...