Inspiration

Solo hosts with 1-3 listings lose 45-90 min per booking rewriting replies as platform policy shifts, and one wrong refund costs payout or a 1-star review.

What it does

StayQuiet runs on a schedule, diffs the policy page clause-by-clause, works only affected bookings, drafts grounded replies and turnover checklists. It pings the host only for refund, exception or review-risk — approve/edit, rest filed quietly with audit trail

How we built it

Strands Agents SDK for Python on Amazon Bedrock (Agent + BedrockModel) with six deterministic tools; Python decides which bookings and whether to interrupt, model decides how to word it. FastAPI + SSE + React quiet monitor, cost guardrail, resilience + golden cache for offline demo.

Challenges we ran into

Fitting 14-message threads to 900 tokens via TS bridge, and making degraded/offline runs still demoable

Accomplishments that we're proud of

Despite the lack of an API to demo a direct connection to AirBnB, the synthetic data replacement of a real API is the only missing part for a fully working application. Each message uses 2 AI agents, and a full python based pipeline. Then it is deterministic but also adds LLMs help in drafting replies.

What we learned

It is possible to benefit from LLMs without adding a chat box. In this case a chat box would have not helped. LLMs giving the already drafted replies works better.

What's next for StayQuiet — Background Agent for Short-Stay Hosts

Real APIs to get input from real bookings servers rather than from synthetic data. A channel for the host to deliver the replies, eg connecting an email server, etc. Adding analytics to each of the properties (each of the 3 listings of the host in the example).

Built With

Share this project:

Updates

Submission history