RepoDoctor

Inspiration: RepoDoctor was inspired by a simple problem: developers can spend hours trying to understand what is wrong with a codebase. I wanted to create a tool that could turn a GitHub repository into a clear, understandable health report instead of overwhelming developers with technical complexity.

How I built it: I built RepoDoctor as a lightweight web application using HTML, CSS, and JavaScript, with GitHub's API for repository analysis and OpenRouter for AI-powered diagnosis. The application retrieves repository information and selected files, performs baseline checks, sends relevant code context to the AI, and presents the results through health scores, findings, evidence, recommended fixes, an X-Ray view, and an AI Quick Help feature. The project is deployed on Vercel with serverless API functions.

What I learned: I learned how to connect multiple APIs into a working AI product, handle repository data safely, design an AI prompt around evidence instead of assumptions, build serverless backend functions, and create a user experience that makes technical results easier to understand.

Challenges: One of the biggest challenges was making the AI analysis reliable within serverless execution limits. I encountered API timeouts and even an invisible Unicode character in an API key that caused a server error. I also had to debug frontend-to-backend communication and reduce the amount of repository data sent to the AI. Solving these problems taught me that building an AI application is not just about getting a model to generate an answer—it is about making the entire system reliable, efficient, and useful.