AI BriefingHow We Built an In-House Data Analysis Agent
Key point
GitHub built an environment where anyone can gain data insights through natural language using Qubot, a Copilot-based data analysis agent.
Details
GitHub developed Qubot, an internal data analysis agent built on Copilot, to solve data accessibility problems that arise in large organizations. Qubot provides answers within seconds when users ask questions in natural language, even without knowing the data model or query syntax.
Qubot's architecture consists of three core components: User Interface, Context Layer, and Query Engine.
- User Interface: Accessible via Slack, VS Code, and Copilot CLI, and on Slack it provides answers to questions in the form of Markdown reports.
- Context Layer: Provides customized knowledge according to the data's stage (Bronze, Silver, Gold). Bronze includes schema information, Silver includes query examples, and Gold includes business rules.
- Context Agent: Reads Markdown-based documentation and normalizes the information into a structured format, continuously enriching the context layer.
This system is not simply a dashboard replacement, but is optimized for Exploratory Questions, such as identifying the cause of retention or metric changes for a specific feature.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.