Why Can't Anyone Answer Questions About the Business?
summary
An internal tool called Studio lets nontechnical teams ask Snowflake Linear and Notion questions in natural language and get reusable outputs rather than one‑off Slack answers. The LLM writes declarative JavaScript widgets that call data sources directly so reruns are deterministic and cheap. Reliability comes from preflight sequencing that injects schema context at tool invocation, a layering rule that forces primary sources over model memory, and query validation that runs every generated Snowflake query before hardcoding. Support teams use the widgets day to day.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:17Business questions internally
- 2:40Introducing Studio
- 4:04How Studio works
- 5:28Reusable tools with widgets
- 7:55Radar support tool example
- 9:39Self-serve dashboards
- 10:01Reliable Studio operation
- 12:51Q&A
related talks
29mKeynote: LLMs, what are they good for?
45mForget SQL! Create your own AI Data Analyst
16mEmpowering Analysts With No-Code Machine Learning
50mMasterclass: Building a data stack in 2021
41mEA built self-service tool Slackbot to provide better user insights
28mBeyond Prototyping: Building Production-Level Apps with Streamlit
19mHow to use generative AI to deliver self-service data products
29mMaking tech boring to keep data exciting