Hostinerary

Hostinerary is a revenue optimization platform for short-term rental hosts, which boosts customer experiences through tailored plans and contributes to local businesses with commission.

Comment

Inspiration

Travel should be fun and stress-free. Hostineraries AI-generated trip plans make stays at Airbnb's unforgettable, and more profitable for both hosts and local attraction businesses.

What it does

Hostinerary gives short-term rental hosts a shareable link they can send to guests. Guests pick their vibe, budget, occasion, and dates, and the app generates a full day-by-day itinerary for their stay using real venues near the property. Then they can chat with an AI concierge to swap activities or ask questions. The host side lets you add local partner businesses (restaurants, spas, tours) with a commission rate. The AI prioritizes those in every itinerary it generates, so hosts earn passive revenue when guests follow recommendations.

How we built it

React + TypeScript frontend, Express API, SQLite. The generation is a two-step pipeline: first we hit OpenAI's Responses API with web_search_preview to find real, rated venues near the property. Then GPT-4o generates a structured JSON itinerary, prioritizing the host's partner businesses over the general results. The chat concierge runs through Backboard, streamed over SSE. Each thread gets the guest's full itinerary injected as context so the AI can handle "can we swap dinner on day 2?" without losing track of the plan.

Challenges we ran into

Getting the AI to use real venue names was the core problem. Vanilla GPT-4o confidently invents restaurants that don't exist. The fix was doing the web search step first and feeding those results into the generation prompt, so the model has actual venue data to work with. Streaming chat through a proxy without buffering issues also took some debugging. X-Accel-Buffering: no and setNoDelay on the socket got it working.

Accomplishments that we're proud of

The commission tracking works end-to-end. When a partner business appears in a generated itinerary, we calculate projected host revenue right there. And the full flow, a 3-day itinerary with real rated venues in about 10 seconds, then chatting to adjust it, feels like something a host would actually send their guests rather than a hackathon demo.

What we learned

Web search grounding is non-negotiable for local recommendations. Without it you're generating plausible-sounding nonsense. The context injection approach for chat (prepending itinerary state to the first message) was simpler than we expected and worked well enough that we'd use it again.

What's next for Hostinerary

Sending the itinerary link automatically when a booking is confirmed via Airbnb/VRBO webhooks. A host analytics view showing which partners get recommended most and actual commission earned. And email/SMS delivery so guests get their itinerary without the host having to copy-paste a link.

Built With

Share this project:

Updates