# VINSHANKS

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Vinod Yallur](https://cerebralvalley.ai/u/vinshanks)
- **GitHub:** https://github.com/vinodyallur/keelcatv12
- **Demo video:** https://youtu.be/JLU4H44nubI
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5098

▶ Try it first: open https://keelcat.in → Sign in → Continue as guest (no account needed). On the Changelog page click "Load sample" (or paste any real API changelog) and watch it become a structured change list. Connect a GitHub repo to get real pull requests.

THE PROBLEM
Modern software runs on other people's APIs: Stripe, Twilio, OpenAI, Anthropic and dozens of internal services. Every one of them changes. A field is renamed, a method is removed, an endpoint is deprecated, and the only notice is a changelog written for humans. You find out when it breaks, usually in production, for a customer, at the worst time. Version bots like Dependabot bump the version number but never rewrite the code that calls the API, so the fix is a manual fire drill: hunt every call site across every repo and language and rewrite it by hand. Around 40% of integration failures come from unmanaged API changes, each costing 15–20 hours of emergency engineering, and AI APIs now deprecate on ~90-day cycles.

I lived this as an intern: a missed changelog took our app down and I shipped 35 pull requests in 9 hours to fix every broken call site by hand. It should have been one reviewed PR, opened automatically. That night is why KeelCat exists.

THE SOLUTION
KeelCat is Dependabot for the APIs your code calls. It watches changelogs, understands the change, finds every affected call site across your repos (TypeScript, JavaScript, Python, Java, C/C++, Go, Ruby), fixes it, verifies the edit and opens one reviewable pull request per repo. A human always reviews; KeelCat never merges on its own.

HOW WE USE META MODELS
Meta Muse Spark 1.3 (meta/muse-spark-1.3) is KeelCat's reasoning layer:
• Understand: it reads messy changelog prose and returns a structured JSON change list (symbol, kind, from → to, impact).
• Fix: it drafts the code edit for changes that need judgment, such as removed or newly required parameters and removed methods.
• KeelCat AI: guests and workspaces without their own key run on an admin-managed Meta model with zero setup and daily limits.
Renames and deprecations are applied by deterministic rules (the model never guesses those), a rule-based parser backs up the model, and every edit is verified before a PR opens. Llama 4 Maverick and Scout, Muse Wonder and local Llama via Ollama are one click away.

BUILT FOR PRODUCTION
Live at keelcat.in with GitHub, Google and guest sign-in, private multi-tenant workspaces, billing and an admin console. Runs execute in the user's browser by default, so repository code never touches our servers. KeelCat currently supports up to 200 concurrent users.

Open source: https://github.com/vinodyallur/keelcatv12

## More from Meta Global AI Developer Hackathon

- [solaoi](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5095)
- [TWINFORGE](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5096)
- [karuna](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5097)
- [mmData](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5099)
- [Jexperiment](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5100)
- [Logosphere BITS (by Ismail Murat Güler)](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5101)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5098. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt