# Clawback

- **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:** [Eyeris Sanchez](https://cerebralvalley.ai/u/Grova)
- **GitHub:** https://github.com/Breistar/clawback
- **Demo video:** https://youtu.be/KXCn8HMAAfI
- **Gallery:** https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/13

Independent hotels lose money to Booking.com and Expedia twice. First, commissions are billed as booked, not as stayed: no-shows, shortened stays and unprocessed corrections get invoiced anyway, and dispute windows close in as little as 48 hours. Second, hotels keep paying 15–20% commission on guests who are already loyal.

Clawback is an autonomous enterprise agent that audits the hotel's entire OTA relationship. It runs three phases in one tool-use loop: a daily SENTINEL that catches unmarked PMS events before they get invoiced (with the contractual countdown on screen), a monthly AUDITOR that reconciles every invoice line against the PMS, the extranet log and each OTA's own contract, and a WIN-BACK engine that identifies repeat OTA guests via RFM and drafts personalized direct-booking offers under the hotel's written benefit ladder.

Every decision cites its evidence (contract clause, PMS record, extranet log, invoice line) with clickable citations and a confidence score — including when the agent declines to act: two reservations with the identical symptom get opposite verdicts, because commission follows the money the hotel actually kept. Correct the agent in chat and it learns: rules persist, totals recompute instantly, and every future audit applies them.

Core reasoning runs on Vultr Serverless Inference (Kimi-K2.6 driving a 13-tool agent loop) and every document retrieval goes through VultronRetriever (rerank), with relevance scores visible live in the reasoning feed. Deployed on a Vultr VPS. Guest-history data is modeled on 12 months of real, anonymized arrival data from the hotel. The problem comes from GROVA's real consulting engagement with an independent hotel in Oaxaca, México; the product was built entirely during the event.

## More from RAISE Summit Hackathon

- [Blue Twilight](https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/10)
- [Justina](https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/11)
- [Hermes](https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/12)
- [Ivan](https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/14)
- [Marshal](https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/15)
- [Bug Catchers](https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/16)

---

Markdown version of https://cerebralvalley.ai/e/raise-summit-hackathon/hackathon/gallery/13. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt