# Finguard

- **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:** [Swayam Shah](https://cerebralvalley.ai/u/0xtigerclaw)
- **GitHub:** https://github.com/0xtigerclaw/finguard
- **Demo video:** https://youtu.be/WMiZfk3mP8E
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1757

A dedicated harness designed for Finance. Inspired by DeepSeek Harness.

Try it: https://finguard-157951168487.asia-south1.run.app

A model can write an investment memo in seconds, and it reads perfectly whether or not the numbers are right. For the last six months I have been building infrastructure like this inside an investment firm. The problem there was the lack of a source of truth.

Finguard takes every number in an investment memo back to the cell in the filing. Muse reads the page images and points to where the filing reports each metric. Plain code reads the PDF's own text there and gives the verdict: match, wrong or refused. An investment memo can be released only if every number matches and a person approves that exact content. The release is sealed with its record and provenance.

Right numbers are the floor. However, more critical is to capture fund's own operations on top, as plugins: house rules, valuation methods and workflows. Describe a method in a sentence, or write one in Python. Every plugin states in code who must approve it and how its result is verified.

## More from Meta Global AI Developer Hackathon

- [kimo](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1754)
- [Attma](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1755)
- [Torpor](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1756)
- [gaphytoro](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1758)
- [SecondCall AI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1759)
- [Sieve- Mask Learning](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1760)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1757. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt