Kin

A private relationship network for hotel guests and the place-makers who serve them. Built for the Rosewood AI Hackathon at Rosewood Sand Hill.

Thesis: Most hospitality AI wants to talk to the guest. Kin is the opposite — AI talks to staff, helps staff talk to guests, and routes context by relationship rather than broadcast.

Tagline: We built the system that lets the people scale.

The two-minute demo

1. Maria (sommelier at Sand Hill) opens her dashboard. Her circle of guests appears, each with a one-line brief in her voice.
2. She opens Apostoli's cockpit. Three role-aware briefs sit side-by-side — same underlying observation, three different abstractions for Maria, Diana (front desk), and Tomás (housekeeping). The sensitive detail never travels to the wrong role. The play button reads the brief aloud (ElevenLabs).
3. Maria taps Capture and dictates a new observation. Pipecat Cloud transcribes; Maître classifies (Haiku 4.5), abstracts per role (Sonnet 4.6), and ghostwrites a message draft in Maria's voice.
4. Maria one-taps Approve. The guest gets an SMS via Twilio with a deep link to a private thread.
5. The guest replies in the thread. Maria sees the reply live via SSE.

Stack

• Next.js 15 (App Router) + TypeScript + Tailwind + radix primitives
• Anthropic SDK — Claude Sonnet 4.6 (drafting + role abstraction) + Haiku 4.5 (classification)
• Prisma + Postgres (Railway)
• Pipecat Cloud dictation agent (separate service in pipecat-agent/)
• Twilio for outbound SMS to the guest's phone
• ElevenLabs for staff-facing audio briefs
• Server-Sent Events for realtime — no third-party signup
• Railway for deploy

Run locally

cp .env.example .env.local
# fill in DATABASE_URL, ANTHROPIC_API_KEY, ELEVENLABS_API_KEY, PIPECAT_API_KEY,
# TWILIO_*, DEMO_GUEST_PHONE, NEXT_PUBLIC_APP_URL

npm install
npx prisma migrate dev --name init
npm run seed
npm run dev

Open http://localhost:3000 — you'll land on Maria's dashboard. The guest view URL is /g/[guestId] (the seed script prints Apostoli's id at the end).

Deploy to Railway

1. New Railway project from this repo
2. Attach a Postgres add-on; copy the DATABASE_URL
3. Set env vars (see .env.example)
4. Deploy. Healthcheck is /api/ping. Migrations run on boot via start-with-migrations.sh.

Deploy the Pipecat dictation agent

See pipecat-agent/README.md. It runs on Pipecat Cloud, not Railway.

What was built during the hackathon

Everything in this repository. The architecture takes inspiration from prior hospitality work, but every file was written fresh for Kin during the event.

Demo controls

• Persona switcher in the staff header (Maria · Diana · Tomás)
• ?as=<slug> switches persona on any staff URL
• ?fallback=1 forces the Type tab in the CaptureModal — backup if Pipecat is sluggish