How We Built Platybot: An AI-Powered Analytics Assistant

Pulumi's development of Platybot, an AI-powered analytics assistant, aimed to address the bottleneck in their #analytics Slack channel by enabling employees to query their Data Warehouse in natural language. To ensure accuracy and reliability, the team built a semantic layer using Cube before integrating AI, which allowed them to define business metrics and prevent errors common with naive AI implementations. Platybot operates as a web app, a Slack bot, and a Model Context Protocol (MCP) server, catering to various user needs from quick lookups to in-depth data analysis. The semantic layer, which organizes data into seven domains, proved crucial as it narrows AI's function to translating user intent into structured Cube queries, ensuring robust and accurate responses. Since its launch, Platybot has processed over 1,700 questions, significantly reducing the data team's workload and allowing them to focus on improving models and data quality. By offering multiple interfaces and transparent reasoning, Platybot has gained user trust and demonstrated the importance of a semantic layer over the AI model itself.