Build an analytics assistant on a semantic layer with web, Slack and MCP access
Pulumi's two person data team, Pablo Seibelt and Lucas Crespo, built Platybot to stop working as a help desk for data requests. They started with a Cube semantic layer over Snowflake, organised into seven business domains, so the assistant works from defined metrics rather than raw tables, with read only SQL as a fallback. Users can choose Gemini 3 Pro, Claude Sonnet 4.5 or Claude Opus 4.6, and reach it through a web app, a Slack bot or an MCP server with six tools launched on February 4, 2026. Since September 2025 it has answered more than 1,700 questions from 83 employees, with daily use growing from 8 to 18 questions by January, when 51 people used it.
Tools used
- Cube
- Snowflake
- Claude Opus
- Gemini
- Slack
- MCP
Sources (1)
- Originalhttps://www.pulumi.com/blog/how-we-built-platybot-an-ai-powered-analytics-assistant/ (opens in a new tab) ↗
Pablo Seibelt and Lucas Crespo (Pulumi) · pulumi.com