TrueStay

Every travel site now has an AI chatbot, but disabled travellers are being failed because the data is superficial. We built an AI agent that calls hotels directly, extracts verified accessibility facts, and provides that dataset to the travel sites and chatbots that are currently guessing.

What it does

A large, verified database of real-world hotel accessibility, built by autonomous phone calls. We find each hotel's number, call the front desk, and ask the questions that matter — shower type, grab bars, accessible parking — extracting structured, categorical answers backed by a verbatim quote and timestamp. We license the dataset directly to travel sites and their chatbots.

How we built it Stay22 for the hotel list Tavily to find phone numbers ElevenLabs Agents Platform + native Twilio integration for the call

Challenges we ran into ElevenLabs' Workflow builder kept erroring — worked around it with a single-agent system prompt instead of a node graph A known end-call bug cuts audio off mid-sentence — worked around by front-loading key words in the closing line Twilio's trial restriction on unverified numbers had to be cleared before real calls would connect

Accomplishments that we're proud of Clean categorical answers plus verbatim quotes, not just a transcript Disclosure and IVR navigation built directly into the agent, not glossed over One schema ties the call script, data collection, and fact sheet together

What we learned Enum-constrained extraction beats parsing free text after the fact Compliance details — disclosure, hold behavior, end-call timing — matter as much as the conversation logic Cutting UI early forced the fact sheet itself to be the polished artifact

What's next for Rich Accessibility Data for Travel Websites Expand to the fuller accessibility question set Turn the fact sheet into a real API travel sites can integrate against Automate the full pipeline end-to-end Pursue direct licensing conversations