We built an internal data analytics agent (github.blog)

GitHub has introduced Qubot, an internal analytics agent powered by GitHub Copilot, to address the long-standing challenge of providing self-serve data analytics across product teams. Traditionally, teams faced significant hurdles in accessing and utilizing product telemetry data, often relying on data analysts to formulate queries and interpret results. Qubot allows GitHub employees to interact with data in plain language, enabling quick answers to exploratory questions about user behavior and product metrics, thereby streamlining workflows and fostering data-driven decision-making. Significantly, Qubot enhances analytics accessibility while functioning as a non-reporting tool that encourages autonomy among employees. Key technical innovations include a three-component architecture comprising a user interface (integrated with Slack and VS Code), a context layer enriched with business-relevant information, and a query engine using both Kusto and Trino for optimized data retrieval. The introduction of an evaluation framework ensures Qubot's responses are accurate and efficient, demonstrating a pivotal shift in analytics engineering. This development not only empowers product and engineering teams at GitHub but also centralizes knowledge, reducing dependency on traditional data analyst roles and promoting a collaborative data culture.