# DepotOps Agent

- **Event:** [RAISE Summit Hackathon](https://cerebralvalley.ai/e/raise-summit-hackathon)
- **When:** Jul 4 at 9:00 AM – Jul 5 at 7:00 PM (GMT+2)
- **Where:** Paris, France
- **Team:** [Yoann Frayce](https://cerebralvalley.ai/u/Yoann)
- **GitHub:** https://github.com/Seveyus/DepotOps-Agent
- **Demo video:** https://www.youtube.com/watch?v=NKzwC8QGDvA
- **Gallery:** https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/104

DepotOps Agent : a live operations agent that knows your night before it happens.

Every evening, an EV(Electric Vehicle) fleet depot makes one high-stakes bet: that tonight's charging plan gets every van to 90% by its departure slot. Late arrivals, a dead charger, a price spike : and the plan that looked fine at 5 PM quietly becomes three missed morning routes. Nobody sees it coming, because nothing looks wrong until dawn.

DepotOps Agent sees it coming. It constructs a live situational model of the depot from streaming events, and after each one it simulates 200 possible remainders of the night (Monte Carlo over a deterministic charging-physics engine). When the failure rate jumps above the depot's measured baseline, not before, it raises a proactive, context-sensitive alert, replays five concrete counteractions across the same 400 sampled futures, and recommends the winner with the numbers to prove it. In the demo: a 50 kW charger dies at 18:40, risk jumps from 20% to 68%, and the recommended action (urgency reprioritization + buffer battery) cuts it to 7% at 12% lower cost on the real EPEX France day-ahead prices of 12 Dec 2024 (CC BY 4.0, fetched and committed during the event).

A non-technical operator can trust it (every claim is a count of concrete simulated outcomes, shown as a dot matrix : "tonight, 200 times"), question it (grounded answers via Crusoe Managed Inference / Nemotron, with a deterministic fallback so the flow never depends on the LLM), and override it (overrides retrain the ranking, visibly). The intelligence lives in the simulation engine; the LLM only writes the sentences : that separation is the architecture, and the engine is the product: swap the parametric uncertainty model for one fitted on a real depot's telemetry (the UncertaintyModel protocol exists for exactly that) and this runs on real money tomorrow.

Built solo, entirely during the event: engine, agent, API, UI, 28 tests : including the scripted demo itself as a CI regression test, Docker, and a live deployment on Azure.

Live demo: http://depotops-agent.francecentral.azurecontainer.io

## More from RAISE Summit Hackathon

- [Athena](https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/101)
- [D-LLM](https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/102)
- [LMT](https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/103)
- [jeremy](https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/105)
- [HQ](https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/106)
- [Wolf-gang](https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/107)

---

Markdown version of https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/104. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt