Inspiration
Post-surgery travel gets treated as logistics, but for the patient it is part of recovery. You leave with mobility restrictions, a medication schedule, and a follow-up appointment, and then you still have to work out where to stay and whether the hotel can actually accommodate you.
Patients should not have to translate aftercare instructions into dozens of travel decisions on their own.
What it does
carestay.ai turns medical recovery instructions into a personalized travel and accommodation plan.
Patients share their instructions however they have them: upload a discharge document, type what the doctor said, or just talk to a voice agent. The agent asks contextual follow-up questions one at a time instead of putting a long form in front of someone who just had surgery.
From that, it extracts the constraints that actually matter (destination, mobility limits, recovery timeline, transportation restrictions, accessibility needs, follow-up appointments) and builds an itinerary with suitable accommodation options.
The shift is from "here are your discharge instructions, figure out the trip" to "here is a recovery-aware plan built around them."
How we built it
We designed it as an autonomous agent with the patient in the loop, not a search form.
Instructions come in through a document, typed text, or conversation. An LLM reasoning layer extracts travel constraints and works out what is still missing, and the voice agent asks for exactly that, one question at a time. Once there is enough, the conversation is converted into structured requirements that drive the itinerary and hotel search.
For the prototype: ElevenLabs for the voice experience, Stay22 for accommodation discovery, Tavily for real-time travel research, and a web interface that brings the instructions, the conversation, the requirements, and the plan into one place.
Challenges we ran into
Turning unstructured medical instructions into travel constraints that matter. A note saying "avoid stairs" or "do not sit for more than two hours" means something specific for hotel selection and transport, and no travel search engine understands that.
Knowing when the agent had enough. We did not want to hand a recovering patient a questionnaire, so the agent had to decide the next most useful question rather than ask everything.
Supporting how instructions actually arrive. Sometimes a PDF, sometimes a verbal explanation, sometimes just what the patient remembers. That pushed us to support all three inputs rather than one.
Accomplishments that we're proud of
We built a continuous workflow rather than a hotel recommender: medical instructions, conversational clarification, travel constraints, accommodation search, itinerary.
We built it around voice, so nobody recovering from surgery has to navigate a complicated interface. And we connected two things normally treated separately, aftercare and travel planning, while keeping the patient in every meaningful decision.
What we learned
The most useful agents are not the ones that decide everything. For sensitive workflows, autonomy works best paired with human input: the AI extracts, researches, and coordinates, while the patient stays the source of truth on comfort and personal constraints.
We also learned how much multimodal input matters. The same instructions arrive as a PDF, a conversation, or a memory, and supporting all three is what makes the product realistic.
What's next for carestay.ai
From planning assistant to full recovery travel coordinator: verifying accommodation accessibility in more detail, coordinating transportation around mobility limits, folding follow-up appointments into the itinerary, and letting caregivers or clinical teams review a plan when appropriate.
Longer term, we want carestay.ai connected to hospitals and outpatient surgery centers so planning can start before discharge. The goal is to own the gap between "you are cleared to go home" and knowing exactly how to get there and recover safely.
Log in or sign up for Devpost to join the conversation.