Inspiration
Hospital readmissions cost the US healthcare system over $26 billion annually — most caused by incomplete discharge instructions, missed follow-ups, and medication errors. We wanted to show how AI agents collaborating with each other (A2A) could catch these gaps before a patient ever walks out the door.
What it does
CareTransit AI takes a patient's clinical profile at discharge and routes it through specialized AI agents — Medication Safety, Follow-up Planner, and Patient Education — only invoking the agents that are clinically relevant, then chains their outputs together into a single, personalized discharge plan.
How we built it
We built a FastAPI orchestrator that implements conditional A2A routing logic, with three specialized agents powered by OpenAI gpt-4o-mini; each agent's output feeds into the next (medication risks inform follow-up urgency, both shape the education content), with full end-to-end observability via LangSmith tracing.
Challenges we ran into
Designing the chaining logic so downstream agents could meaningfully consume upstream outputs — not just pass raw text, but inject structured context — required careful prompt engineering to keep responses coherent and clinically grounded.
Accomplishments that we're proud of
We built a working multi-agent A2A system where agents genuinely influence each other's reasoning, complete with a visual frontend that shows routing decisions, chain arrows, and the final synthesized plan — all in a single weekend.
What we learned
A2A architecture isn't just about parallelism — the real power is in selective routing and context chaining, where knowing when not to run an agent is just as important as knowing what to do when you do.
What's next for CareTransit AI
We plan to add EHR integration (FHIR API), a feedback loop where clinicians can flag agent recommendations to retrain routing thresholds, and expand to additional specialized agents for social determinants of health and insurance pre-authorization.
Log in or sign up for Devpost to join the conversation.