Chowkidaar

Chowkidaar is an autonomous AI watchman that predicts API failures, maps the exact code at risk, and generates tested pull requests before production breaks.

Comment

Inspiration

Modern apps depend on dozens of external APIs, but developers usually discover breaking changes only after something fails. I wanted to build a “watchman” for API-powered software: an agent that understands where APIs live inside a repo, detects risk early, and prepares a verified fix before production breaks.

What it does

Chowkidaar connects to a local or GitHub repository, builds a visual dependency graph of API-related code, watches for API drift, environment changes, and traffic pressure, then recommends targeted fixes. It can simulate peak load, identify bottlenecks, write a focused patch, run project checks, and show a PR-ready repair workflow.

How I built it

I built a React frontend for the graph UI, simulation panel, audit workflow, and PR timeline. The backend uses FastAPI-style services for repo connection, graph generation, API scanning, simulation, audits, and repair orchestration. The system slices the repository into API pipelines so the agent works on only the relevant files instead of the whole codebase.

Challenges I ran into

The hardest part was making the product feel like an actual agent instead of a static dashboard. I had to connect local folders, GitHub repos, graph rendering, simulations, audits, and the Apply workflow into one smooth experience. Another challenge was keeping the AI workflow safe: the agent should be helpful, but developers must stay in control.

Accomplishments that I'm proud of

I am proud that Chowkidaar feels end-to-end: connect a repo, see the API graph, run a simulation, get a recommendation, apply it, run checks, and reach a PR-style review screen. I also built the interface to explain why a change is recommended, not just produce code.

What I learned

I learned that AI coding tools become much more powerful when they are grounded in structured context. A graph of the codebase, traffic signals, and validation checks makes the agent more useful than a generic chatbot. I also learned how important UX is for trust: developers need to see evidence, not magic.

What's next for Chowkidaar

Next, I want to deepen real GitHub PR automation, expand provider-specific API drift detection, improve live traffic reporting, and add stronger review-comment handling. Long term, Chowkidaar could become a continuous API reliability layer for teams that depend on fast-changing external services.

Built With

Share this project:

Updates

Submission history