Hospitable Buildathon 2026

The AI Concierge That
Books Experiences, Not Just Answers

An intelligent agent that reads guest messages, detects intent, curates premium local experiences — and actually executes the bookings. Hosts earn commissions. Guests get magic.

90s
Avg. Response Time
$120
Avg. Commission / Stay
4.9★
Partner Minimum
0
Bad Experience Tolerance
01 — System Architecture

End-to-End Message Flow

From guest inquiry to confirmed booking — every step orchestrated, every partner vetted, every experience guaranteed.

🧳 Guest Sends message 📡 Hospitable MCP Webhook trigger 🤖 AI Concierge OpenClaw + Claude 🧠 Context Engine Trip type · Party size · Dates 💬 Response Gen Curated suggestions 📤 Send to Guest via Hospitable MCP ✅ Guest Says YES "Book that dinner!" ⚡ Execution Layer Route to correct booking channel 🍽 Restaurant Booking OpenTable · Resy · Email ⛵ Yacht Charter Email to vetted captains 📸 Photographer Email + scheduling 🎯 Activities Viator · GetYourGuide ✓ Confirmation Details sent to guest Commission tracked
🔁 Feedback Loop

After every booking, the agent sends a follow-up message to the guest asking about their experience. Negative feedback triggers an immediate host notification — intercepting bad reviews before they happen.

02 — Message Intelligence

Context Extraction Engine

The agent doesn't just respond — it understands. Every guest message is parsed for trip intent, party composition, timing, and emotional tone.

🎯
Trip Type Detection

Claude analyzes the booking message + any follow-up messages to classify the trip into one of six categories, each with tailored recommendation profiles:

🎉 Celebration 🥂 Bachelorette ❤️ Anniversary 👨‍👩‍👧‍👦 Family 💼 Business 🧘 Solo Retreat

Each type maps to a curated set of suggestions. Bachelorette? Yacht charter + rooftop dinner. Anniversary? Sunset photography + fine dining. Family? Kid-friendly activities + casual restaurants.

📊
Party Size & Timing

Extracted from the Hospitable booking data and guest messages:

  • Party size — determines restaurant table size, charter vessel, group discounts
  • Check-in / check-out dates — scopes all suggestions to the stay window
  • Day-of-week awareness — weekend vs. weekday availability, brunch vs. dinner
  • Season/event awareness — local festivals, peak pricing, limited availability
  • Lead time — same-day requests trigger quick-book partners vs. advance reservations
🎭
Tone Matching

The agent mirrors the guest's communication style:

  • Casual guest → warm, emoji-friendly, conversational
  • Formal guest → polished, concise, professional
  • Excited guest → high energy, matching enthusiasm
  • Brief/direct → short responses, no fluff

This isn't just politeness — it's conversion optimization. Guests are more likely to book when the response feels human and matches their vibe.

Suggestion Mapping

Every suggestion is selected from the vetted partner list based on the extracted context:

  • Trip type → filters partner categories
  • Party size → filters capacity-compatible options
  • Budget signals → inferred from property tier and guest language
  • Past guest data → what similar guests loved (aggregated, anonymous)

Suggestions are presented as 2–3 curated picks, never a dump of 10 options. Decision fatigue kills conversion.

💡
Example: Message → Intelligence → Response
Incoming Message
"Hey! So excited — it's my best friend's bachelorette weekend! There will be 8 of us arriving Friday. Any restaurant recommendations for a fun dinner Saturday night? We're also interested in a boat day if that's a thing 😍"
Extracted Context
Trip Type: Bachelorette
Party Size: 8
Key Date: Saturday dinner
Interests: Dining, Boat charter
Tone: Excited, casual
Budget Signal: Mid-high (emoji use, "fun")
Generated Response
"Omg how fun!! 🎉 You're going to LOVE San Diego. For Saturday dinner, I'd recommend Coasterra — stunning waterfront views, great cocktails, and they have a semi-private area perfect for a group of 8. For the boat day, Captain Mike runs an amazing charter — I can get you a half-day on Sunday with drinks included. Want me to book both?"
03 — Execution Layer

From "Yes!" to Booked

When a guest confirms interest, the agent doesn't just give a link — it executes the booking. Each service type has its own booking pipeline.

🍽
Restaurant Booking

1. API Booking (Primary)

Attempt reservation via OpenTable API or Resy API with party size, date, time, and special requests.

2. Email Fallback

If no API integration exists for the venue, send a templated email to the restaurant's reservation email with all details.

3. Confirmation

Monitor for booking confirmation (API callback or email reply via IMAP). Relay confirmation with details + referral link back to guest.

API-first Email fallback Commission tracked
Yacht & Charter Booking

1. Partner Email Outreach

No booking APIs exist. Agent sends a structured availability request to pre-vetted charter captains via SendGrid/Resend.

2. Reply Detection

Gmail IMAP polling watches for captain replies. Agent parses availability, pricing, and vessel options.

3. Present Options

Summarize 1–2 best options for guest. On confirmation, agent replies to captain with booking confirmation.

Email-only IMAP polling Manual handoff available
📸
Photography Sessions

1. Availability Check

Email vetted photographers with date, location preference, session type (golden hour, event, portrait).

2. Scheduling

Coordinate timing with guest itinerary. Suggest optimal times (golden hour for beach, etc.).

3. Delivery Tracking

Follow up with photographer post-session to ensure <72hr delivery to guest. Flag delays to host.

Email outreach Scheduling Delivery SLA
🎯
Activities & Experiences

1. API Search

Query Viator or GetYourGuide APIs for activities matching trip type, party size, and dates.

2. Filter & Rank

Apply vetting filters (rating, reviews, cancellation policy). Rank by relevance to guest context.

3. Book & Confirm

Execute booking via API. Send confirmation with meeting point, times, and what to bring.

API booking Auto-filtered Affiliate links

Commission Tracking

🧳
Guest Books
🔗
Unique Referral Link
🍽
Partner Confirms
💰
Commission Logged
📊
SQLite Tracking

Every booking generates a unique referral link. Partner reports commission completion. Host dashboard tracks earnings per guest, per partner, and per month.

04 — Reputation Protection

Partner Vetting System

This is the most critical system. The host's Airbnb rating — years of 5-star reviews — depends on every recommended experience being excellent. No exceptions.

⚠️ The Core Principle

The agent only recommends partners from the approved list. Every partner is manually vetted by the host before being added. The AI never independently discovers and recommends a business. The vetting criteria below are the minimum bar — the host has final say.

🛡
Vetting Criteria by Category
Category Min Rating Min Reviews Additional Requirements
🍽 Restaurants 4.2★ Google 200+ No health code violations. Agent monitors live rating — auto-flags drops below threshold.
⛵ Yacht Charters 4.5★ 50+ USCG licensed. Liability insurance verified. Safety record checked.
📸 Photographers 4.5★ 30+ Portfolio review by host. <72hr delivery guarantee. 3 references checked.
🎯 Activities 4.3★ 100+ Free cancellation preferred. Insurance verified for adventure activities.
📡
Continuous Monitoring

Live Rating Watch

Agent checks Google review scores for all partners weekly via Google Places API. Score drop below threshold → auto-flag to host.

Guest Follow-Up

After each booking, agent messages guest: "How was dinner at Coasterra?" Negative feedback → host notified immediately.

Health & License Checks

Quarterly verification of restaurant health inspections, charter USCG licensing, and insurance status.

🚨
Three-Strike System

Protecting the host's reputation requires zero tolerance for consistent failures:

1
Strike 1: Negative guest feedback logged. Partner notified. Host alerted.
2
Strike 2: Partner deprioritized in recommendations. Host reviews relationship.
3
Strike 3: Partner auto-removed from approved list. Agent stops recommending immediately.
🎯 The Goal

Intercept bad experiences BEFORE they become Airbnb reviews. The follow-up message catches problems in real-time, allowing the host to intervene, apologize, and offer alternatives — turning a potential 3-star review into a 5-star recovery story.

05 — Technical Stack

Built With

A lean, modern stack designed for a hackathon but architected for production. Every component chosen for reliability and speed.

📡
Hospitable MCP
Guest messages in/out
🧠
Claude API
Context extraction & response gen
🦞
OpenClaw
Orchestration layer
🍽
OpenTable + Resy
Restaurant booking APIs
🎯
Viator / GYG
Activity booking APIs
📧
SendGrid / Resend
Email outreach to partners
📬
Gmail IMAP
Reply detection & parsing
🗄
SQLite
Partners · Guests · Commissions
Cloudflare Workers
Edge hosting & webhooks
🔗
Referral Links
Commission attribution
📍
Google Places API
Live rating monitoring
🔔
Push Notifications
Host alerts & escalations
🏗
System Architecture
INPUT LAYER Hospitable MCP Gmail IMAP Webhook Events PROCESSING LAYER OpenClaw Agent Orchestration Claude API NLU + Generation SQLite DB State + History EXECUTION LAYER OpenTable Resy SendGrid Viator GYG MONITORING Google Places Rating Watch · Guest Follow-up Loop · Partner Health Checks · Commission Reconciliation
06 — Hackathon Demo

Loom Video Flow

A 3–5 minute demo that tells a story. Not a feature tour — a narrative that makes judges feel the magic.

1
0:00 – 0:30

The Problem

Quick hook: "Every Airbnb host gets the same question — 'Any restaurant recommendations?' — and every host copy-pastes the same list. What if the house itself could book the dinner?" Show a typical generic Airbnb message exchange. Boring, manual, no revenue.

2
0:30 – 1:00

The Solution — Live Demo Setup

Show the Hospitable dashboard briefly. "We built an AI concierge that reads guest messages, understands what they actually want, and books it for them." Show the OpenClaw agent config.

3
1:00 – 2:30

Live Message → Response → Booking

The money shot. Send a test guest message: "Hey! It's my wife's birthday, party of 4, any nice dinner spots Saturday?" Show the agent detecting anniversary/celebration, party of 4, Saturday dinner. Watch it generate a warm, personalized response with 2 restaurant picks. Simulate guest saying "Book Coasterra!" → watch the agent execute the OpenTable API call → confirmation sent back.

4
2:30 – 3:30

The Trust Layer

This is the differentiator. Show the partner vetting dashboard: rating thresholds, three-strike system, the follow-up message flow. Emphasize: "We don't just recommend — we guarantee. Every partner is vetted. Every experience is tracked. Negative feedback triggers an intervention before it becomes a review."

5
3:30 – 4:00

The Business Model

Show the commission tracking: unique referral links, per-guest revenue, monthly earnings dashboard. "The host earns $50–200 per stay in commissions — from experiences they didn't lift a finger to book." Flash the numbers: projected revenue per property per year.

6
4:00 – 4:30

Close — Why This Wins

Three points: (1) Built on Hospitable MCP — real integration, not a mockup. (2) Reputation protection is baked in — not bolted on. (3) Revenue generation for hosts — this pays for itself day one. End with: "The best concierge doesn't wait to be asked. It anticipates."

💡 Demo Tips

Pre-record the API calls to avoid live demo failures. Have a real guest message thread ready in Hospitable. Show the actual code briefly but don't dwell — judges care about the product, not the implementation details. Keep the energy up — this is a pitch, not a lecture.