Implementing Natural Language Analytics with Snowflake Cortex
summary
GitLab implements conversational analytics with Snowflake Cortex and a simple Streamlit front end, starting with a 3–5 table POC and evolving through prompt engineering and a governed semantic model (YAML) that maps business terms to tables. Accuracy for simple queries rises from ~60% to 85–95% and complex queries from ~40% to ~75%; People Ops request volume drops about 50%. Data never leaves Snowflake, role‑based access, verified‑query feedback, Atlan metadata integration, and monitoring address governance, ambiguity, and trust issues.
AI-written summary · about
chapters (12)
- 0:00Natural language analytics intro
- 3:08Limitations of BI dashboards
- 5:12BI limits solved with Cortex
- 9:31Cortex Analyst features
- 12:44Semantic model building
- 19:04Pilot proof of concept
- 21:51Improving accuracy and speed
- 23:33Edge cases and security
- 27:26Business adoption phases
- 28:49Key lessons learned
- 30:35Outcomes and benefits
- 31:25Q&A
related talks
33mIntegrating Serverless LLMs with Data Platforms for Natural Language Analytics
28mAutomating Database Operations with Cortex Code
49mLLM inference and fine-tuning on secure enterprise data
27mUsing GenAI with Audio Data for Driving Better Customer Experience
23mAcceleratinag Data Science and Machine Learning with Snowflake DataTech
26mGraph Analytics in Snowflake: Beyond Traditional SQL
15mFrom Air Quality to Aircraft and Automobiles, Data Is Everywhere
28mUsing Dataops To Build And Manage Data Products On Snowflake