Building DataBot: Our always-on data assistant

Sourcegraph developed DataBot, an AI-powered Slack bot designed to minimize context-switching for data analysts by addressing routine data inquiries, allowing the data team to focus on auditing rather than performing analyses. DataBot operates by leveraging various tools and custom data warehouse contexts to answer questions about user behavior, telemetry, and product metrics without human intervention, thus streamlining the process for quick questions that would otherwise disrupt analysts' deep work. The bot is equipped with a suite of tools, including BigQuery, PostHog, and Looker, and is guided by SCHEMA_CONTEXT, which helps it understand the business logic and data warehouse structure, ensuring accurate responses. By maintaining thread context, DataBot can handle follow-up questions seamlessly, building on previous answers for a cohesive dialogue. This shift allows data analysts to audit responses quickly, akin to reviewing pull requests, and drastically reduces the time spent on each query, as DataBot's capabilities improve with each interaction.