# AgentRose

- **Event:** [Hospitality 2030: A Rosewood Sand Hill Hackathon](https://cerebralvalley.ai/e/rosewood-hospitality-2030)
- **When:** Sat, May 16 at 9:00 AM – 10:00 PM (PDT)
- **Where:** Rosewood Sand Hill, Menlo Park, CA
- **Team:** [Stanley Mei](https://cerebralvalley.ai/u/SCU), [Eric Nelson](https://cerebralvalley.ai/u/stylo_eric), [Faraz Fazel](https://cerebralvalley.ai/u/Faraz23), [Annie Ostojic](https://cerebralvalley.ai/u/AnnieO)
- **GitHub:** https://github.com/agentrose2026/agentrose2026/tree/main/backend/app/seed
- **Demo video:** https://youtube.com/shorts/3pPTPjM8oOM?feature=share
- **Gallery:** https://cerebralvalley.ai/e/rosewood-hospitality-2030/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/rosewood-hospitality-2030/hackathon/gallery/46

AgentRose — an AI concierge that turns guest text messages into completed resort services, with humans supervising
  only the edge cases.

  Built for Rosewood Sand Hill, a 121-room luxury resort in Menlo Park, AgentRose runs the full guest-request loop end
  to end:

  - Proposes — reaches out to a guest over text with a specific offer (a dinner reservation, a spa slot, a car, a fix)
  and tracks it as a live, time-limited offer.
  - Understands the reply — reads the guest's response, decides whether it's a yes, a no, or a question, and answers in
  the resort's voice.
  - Closes the loop — on a yes, the service is booked and confirmed automatically; on anything unclear, it hands off to
  staff rather than guessing. AgentRose decides, but a guardrail keeps it from ever booking on an ambiguous reply.
  - Learns the guest — pulls preferences, allergies, party details, and milestones out of conversations into a single
  Guest 360 profile, so every future interaction is more personalized.

  This sits on top of a four-stream service operation — Food (restaurant, bar, in-room dining), Spa, Concierge
  (experiences & transport), and IT — each with its own work queue, SLA tracking, cross-team handoffs, and role-based
  access. The net effect: guests interact entirely by text, AgentRose handles the negotiation and confirmation, and
  staff focus on exceptions instead of data entry. A staff-facing AI copilot also drafts suggested replies grounded in
  the resort's knowledge base.

  Technical footing: a FastAPI/Postgres backend with a Next.js/TypeScript frontend, Claude powering the agent and
  copilot, and SMS delivery via Sendblue. Every AI feature degrades gracefully — it runs on deterministic logic with no
  API key and upgrades to Claude automatically when one is present, so the product always demos and never hard-fails.

---

Markdown version of https://cerebralvalley.ai/e/rosewood-hospitality-2030/hackathon/gallery/46. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt