# SOVIR

- **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:** [Helmut Chaparro Sandoval](https://cerebralvalley.ai/u/hchaps404)
- **GitHub:** https://github.com/HCHAPS404/SOVIR
- **Demo video:** https://youtu.be/PdG3e74f0Fs
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4917

SOVIR protects people from manipulative and fraudulent digital environments: scam funnels, credential phishing, coercive ads, fake urgency, dark patterns. It works where these threats actually live, in the browser, on the phone, and in a web console, instead of asking the user to forward suspicious messages to some separate tool.

The problem is that people make digital decisions inside interfaces designed by third parties. Risk arrives fragmented across ads, SMS, websites, apps and recommendation feeds, and existing tools either watch a single channel or stay blind until the damage is done. Nobody gives you a contextual, verifiable answer to a simple question: what is being asked of me right now, and how dangerous is it?

SOVIR answers that question without taking your data to do it. With specific, revocable consent, it observes screen metadata only, never raw text, passwords or screenshots, and emits typed signals like "this screen requests payment, creates urgency, pushes you off-platform." A deterministic policy engine fuses those signals into interventions the user can read and verify. Facts, hypotheses and actions stay strictly separated at the type level. Models propose; deterministic policy decides.

I built SOVIR from zero for the indie track, but I am not treating it as a hackathon project. The repo already runs as a local-production scaffold: Rust hexagonal core, Postgres with row-level security, real OIDC auth, a Chrome extension, an Android accessibility observer, a Next.js console and a public site. Privacy is enforced by architecture, not by a privacy policy. My intent is to keep building it past the competition and grow it into a business: the consent model, the typed-signal taxonomy and the verifiable trust boundary are a product foundation, not a demo. The roadmap already contemplates what a company needs, from domain-specific protection packs and team administration to a paid tier with remote-analysis budgets. I want SOVIR to become a real privacy company, one where the protection people receive is something they can verify, not just believe.

## More from Meta Global AI Developer Hackathon

- [Everything Everywhere All at Once](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4914)
- [Kenosis Interactive](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4915)
- [Open Book Studio](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4916)
- [People Organised - Marx de Morais](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4918)
- [Team Shamas](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4919)
- [Jermaine Dennis](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4920)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4917. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt