Software repositories store massive amounts of code contributions, bug reports, and project activity data, yet non‑technical stakeholders and developers often lack the expertise to query these sources effectively. To tackle this, we propose a novel chatbot architecture that leverages OpenAI's GPT‑4 model together with a tool‑selection mechanism for automated extraction and analysis of repository data. The workflow consists of three stages: first, parse the user query to extract relevant parameters; second, match those parameters to an appropriate tool (e.g., Issue search, Pull Request lookup, Commit listing); third, feed the tool's raw output into GPT‑4 as context to generate a detailed, actionable response. Unlike prior multi‑component systems that rely on embedding models and document retrievers, our design uses prompt engineering and intent‑driven tool selection, eliminating semantic drift during retrieval. Experiments span five question categories—Issues, Pull Requests, Commits, compound questions, and general repository information—and assess how tailored prompts improve the model's answer accuracy. Findings demonstrate that the architecture not only boosts response correctness but also lowers the barrier for non‑technical users to obtain actionable repository insights, serving a broad audience.
Review