# Strawhat

- **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:** [Mihir Panchal](https://cerebralvalley.ai/u/mihir2603)
- **GitHub:** https://github.com/sLaYerUnleAsheD/aegis-migrate
- **Demo video:** https://www.youtube.com/watch?v=ROevSqyPVKw
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1835

Roughly $3 trillion in daily commerce flows through COBOL systems, including around 95% of all ATM transactions. However, the average COBOL programmer is 58 years old, and these critical developers are retiring at a rate of roughly 10% per year without being replaced. This leaves the global financial system facing an impending crisis: hundreds of billions of lines of legacy code with nobody left to maintain them. While enterprises desperately need to modernize this infrastructure, manual migrations take years and cost hundreds of millions of dollars, and standard AI code generators cannot be blindly trusted with mission-critical banking logic.

An LLM will happily translate COBOL, and the result reads correctly. But COBOL has behaviors that look harmless and aren’t: implied decimal points, ROUNDED versus truncation, zero-suppressed number formatting, and scope terminators that quietly change which statements belong to a branch. Reading the code won’t find those. Running both programs and comparing their output will.

Aegis Migrate translates legacy COBOL batch programs to Go and checks the result. It compiles and runs the original COBOL, runs the Go translation on the same generated inputs inside a no-network Docker sandbox, and compares every output byte. If they differ, the compiler error or output diff goes back to the model, which fixes its code (up to three repairs). I ran it on 12 COBOL programs across Meta’s models, over 400 runs in all, and built a results viewer at https://aegis-migrate.netlify.app/ where each run can be replayed.

## More from Meta Global AI Developer Hackathon

- [SightWork](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1832)
- [OG](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1833)
- [EventOps](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1834)
- [Futurial](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1836)
- [BTECH VPSE-Rx // 3D VOLUMETRIC DOSING & COUNTERFEIT INTERLOCK](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1837)
- [Nikhil](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1838)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1835. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt