The Rise of Qubot: GitHub’s AI-Powered Analytics Game-Changer
In the fast-paced world of software engineering, accessing and interpreting data quickly can make or break decision-making. At GitHub, Matteo Vasirani, the staff manager of software engineering, faced a familiar challenge: making data analytics truly self-serve for dozens of product teams. Despite years of industry efforts, this problem has remained largely unsolved—until now. With the help of AI, GitHub has found a credible solution in the form of Qubot, an internal analytics agent powered by GitHub Copilot.
What is Qubot?
Qubot is not your typical reporting tool or dashboard replacement. Instead, it’s designed for exploratory analytics. Imagine being able to ask questions like, “Which cohort of users has the highest retention on this feature?” or “What product contributed most to this metric last week?” in plain language and getting answers within seconds. That’s exactly what Qubot does. It’s accessible, intuitive, and requires zero maintenance, making it an invaluable tool for teams unfamiliar with complex datasets.
How Qubot Works
Qubot’s architecture is divided into three main components:
-
User Interface: Qubot is available through Slack, VS Code, and the Copilot CLI. Slack is the go-to collaboration tool for GitHub employees (affectionately called Hubbers), and it requires no configuration. Questions posted in the Qubot Slack channel trigger a Copilot Cloud Agent, which provides answers directly in the thread. Results are also stored as markdown reports in a pull request for further refinement.
-
Context Layer: GitHub’s data warehouse contains data at various curation stages—raw events (bronze), conformed facts and dimensions (silver), and curated datasets (gold). The context layer is built in a federated way, with knowledge tailored to the type of data. For example:
- Bronze data: Includes telemetry context from product teams.
- Silver data: Includes query examples and usage guidance from the data and analytics team.
- Gold data: Includes business rules and metric definitions from dataset-owning teams.
The context layer is enriched with signals from ETL pipelines and is loaded at runtime via the GitHub MCP Server. This federated approach ensures that the context remains up-to-date and relevant.
-
Query Engine: Qubot connects to both Kusto and Trino, the two query engines powering GitHub’s analytics. Kusto is ideal for exploratory questions over recent data, while Trino handles complex joins and deeper historical analysis. Qubot defaults to Kusto but switches to Trino when necessary, ensuring users get the best performance without needing to know which engine to use.
The Impact of Qubot
Qubot has been a game-changer at GitHub. With hundreds of users running thousands of queries, it has dramatically reduced the need for data analysts to field questions in Slack channels. Teams can now explore data with greater autonomy, and even those who were hesitant to dive into the data warehouse can access the insights they need.
One of the key lessons learned is the importance of the context layer. A well-curated context not only makes Qubot more accurate but also three times faster at returning answers. This highlights the need for structured, high-quality context in AI-driven analytics tools.
Qubot’s success also demonstrates the power of a hub-and-spoke model. By centralizing distributed knowledge into a single tool, Qubot has removed strain from the data and analytics team while incentivizing other teams to contribute. Product teams own their telemetry, and business teams own their gold data, creating a collaborative ecosystem that benefits all of GitHub.
The Future of Data Analytics at GitHub
Qubot is more than just a tool—it’s a step toward democratizing data access and empowering teams to make data-driven decisions. By bridging the gap between complex datasets and user-friendly queries, Qubot is paving the way for a more data-literate workforce. As AI continues to evolve, tools like Qubot will become increasingly essential in unlocking the full potential of data.
Final Thoughts
GitHub’s Qubot is a testament to the power of AI in transforming how we interact with data. By making analytics accessible, intuitive, and fast, Qubot is setting a new standard for data exploration. Whether you’re a seasoned data analyst or a curious team member, Qubot ensures that the insights you need are just a question away.
For more on GitHub’s innovations, stay tuned to our blog! And if you’re interested in diving deeper into AI and machine learning, check out our related posts.