# Solo Yolo

- **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:** [Sarthak Sarthak](https://cerebralvalley.ai/u/Sarthak99106)
- **GitHub:** https://github.com/sarth99106/gatekeeper-mongodb-agent.git
- **Demo video:** https://youtu.be/n6S5lf6i_Ks
- **Gallery:** https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/17

Gatekeeper is an auditable financial operations (FinOps) AI agent that moves authorization, policy enforcement, and auditability out of fragile LLM prompts and into the database layer. Built with MongoDB Atlas, LangChain (Groq), Voyage AI, and ElevenLabs, Gatekeeper ensures that autonomous customer-support and billing agents can never execute unauthorized, over-cap, or unverified financial mutations.

When a customer support ticket or account action is submitted, LangChain and Groq extract structured intent using strict Zod schemas. Before any customer balance changes, Gatekeeper enforces two layers of protection:

Human-in-the-Loop (HITL) & Identity Guardrails: Automatically pauses execution to require operator confirmation and mandatory audit reasoning on customer deductions (CHARGE_CUSTOMER), duplicate-charge amount discrepancies, unrecognized customer names, and account ID mismatches.

Pre-Execution Policy Gate & Escalation Workflow: Enforces role-based permissions (support_tier_1, PIN-protected finance_lead, and untrusted_bot). Over-cap requests and new account onboardings from Tier-1 support are blocked before execution and routed to a PIN-protected finance_lead approval inbox.

Every approved, blocked, escalated, and rolled-back action is recorded in an immutable audit ledger capturing who authorized it, what changed (pre_state to post_state balance deltas), and why (both LLM reasoning and operator audit notes). Operators can query the audit trail using hybrid natural-language and numeric search, trigger an emergency session kill-switch, or selectively roll back individual or batched transactions with mandatory rollback reasons and spoken ElevenLabs voice narration.

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

- [Readback](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/14)
- [Countersign](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/15)
- [PromptForge](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/16)
- [Ontara](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/18)
- [Refund guard](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/19)
- [Bedley](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/20)

---

Markdown version of https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/17. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt