Answer data questions in Slack with a text to SQL agent and confidence signals

Fullscript built Flurry, a Slack agent that turns plain questions into SQL so staff can get data without waiting on analysts. A prototype came together in about a week using Claude Code with Claude 4.5 Opus, and executives piloted it in December. The production version uses Google ADK with Gemini 3 Flash over Snowflake semantic models and a library of verified queries. Each answer carries a high, medium or low confidence signal, PII and PHI detection guards sensitive data, and LLM judges score live traffic with results logged to Langfuse. Flurry now handles about 800 queries a week from more than 50 users.