# FixLoop

- **Event:** [Agents in Context Toronto: a MongoDB Hackathon](https://cerebralvalley.ai/e/mongodb-toronto-hackathon)
- **When:** Sat, Nov 14 at 9:00 AM – 9:00 PM (EST)
- **Where:** Toronto, Canada
- **Team:** [Kwabena kesse](https://cerebralvalley.ai/u/skypto)
- **GitHub:** https://github.com/skypto/fixloop
- **Demo video:** https://youtu.be/cNoFj-vUJYQ
- **Gallery:** https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/27

FixLoop prevents apartment maintenance teams from repeating failed repairs by turning technician field notes into verified, human-governed AI memory.

How It Works: The 3-Step Workflow
Investigate: A technician logs a recurring issue (like a ceiling stain), gets an AI-assisted inspection recommendation, and records what actually fixed it. FixLoop drafts a reusable lesson tied directly to that work order.

Verify: The AI cannot learn on its own. Every drafted lesson sits in a pending state until a human supervisor reviews the evidence and approves where it applies (building scope), who can access it (roles), and when it expires.

Prove: When a similar complaint comes in later, even if worded differently, FixLoop runs a side-by-side comparison of the AI with Memory OFF versus Memory ON. Using MongoDB Atlas Vector Search, the memory-enabled AI retrieves only approved, unexpired lessons for that building, cites the exact work order, and shows judges honestly whether that memory changed the recommended action.

Built-In Governance & Trust
The Memory Trust Panel: Gives supervisors full visibility into every lesson's source, approval status, scope, and expiration date.

Instant Revocation & Forgetting: Supervisors can revoke a lesson or wipe its AI embeddings at any time while keeping the original work orders and audit logs completely intact.

Strict Backend Guardrails: Code-level policy checks verify permissions and expiration dates before and after every search, ensuring revoked or outdated data never leaks into a new recommendation.

## More from Agents in Context Toronto: a MongoDB Hackathon

- [CONTEXT RETAIL : The Governed Memory Platform for AI-Powered Retail Operations](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/24)
- [suture](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/25)
- [Yes_chef](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/26)
- [Precedent](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/28)
- [AWP](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/29)
- [Lore](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/30)

---

Markdown version of https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/27. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt