# LA Hacks

- **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:** [Harshi Kamboj](https://cerebralvalley.ai/u/harshikamboj), [Divjot Singh Kamboj](https://cerebralvalley.ai/u/Divjot), [Keerat Ghuman](https://cerebralvalley.ai/u/KeeratGhuman), [Navdeep Rajpal](https://cerebralvalley.ai/u/navdeep24)
- **GitHub:** https://github.com/harshikamboj/LAHacks
- **Demo video:** https://drive.google.com/file/d/1yUr5SdNPMuwlXq1Mgdq4Xl0hC2Z6xV2R/view?usp=drive_link
- **Gallery:** https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/88

TradeOps is a financial reconciliation application that demonstrates ProofStep, our control-layer architecture for reliably completing multi-step workflows across independent systems. Built for Track 3, “Finish the Job,” it addresses a critical operational problem: when a response is lost, processing crashes, or an upstream system changes, how do you determine what actually happened and finish the task without duplicating an action?

TradeOps investigates rejected settlement instructions by comparing broker trade records, receiving-account directories, and custodian instructions. It validates the proposed correction and requests human approval for the exact change. Before submitting anything, ProofStep persists the approved intent and a stable operation identifier in MongoDB Atlas. After submission, it independently checks both the custodian’s receipt and the updated instruction before marking the correction complete. Our demo includes a real worker crash after the custodian records a correction but before its response is received. A restarted worker resumes from the saved checkpoint and verifies the original operation, with one recorded correction and one submission.

Verified outcomes also improve subsequent investigations. TradeOps saves successful, independently confirmed procedures as experience, generates real Voyage AI embeddings, and retrieves relevant experience through Atlas Vector Search. A similar follow-up case demonstrates that this memory changes behavior: the investigator proactively checks the live directory format before reading the receiving account. Current records are still validated, and every correction requires its own approval.

The operational interface presents an actionable work queue covering account corrections, similar follow-up incidents, already-accepted instructions requiring no amendment, and conflicting records requiring specialist review. It incorporates 66,693 public SEC fails-to-deliver observations as historical market context, while trades, accounts, instructions, and corrections are clearly simulated.

TradeOps is the application users operate; ProofStep supplies the durable state, approval gates, recovery discipline, verified completion, and reusable experience underneath it. The demonstrated outcome is custodian acceptance for processing, rather than final securities settlement.

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

- [Lyfe Events](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/85)
- [Miro](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/86)
- [Dxp](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/87)
- [TestEcho](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/89)
- [sential](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/90)
- [Gomon](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/91)

---

Markdown version of https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/88. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt