AI Triage Assistant

Patient portal home
Nurse portal
Patient portal voice
Patient portal text

Inspiration

In emergency departments, the first triage decision is often made through a short conversation, before vital signs or tests are available. Some urgent conditions do not look serious at first, making them harder to identify early.

This made us ask a simple question: How can we support this first triage conversation so that hidden urgency is less likely to be missed?

We wanted to explore how AI could support this early moment by listening more carefully and consistently, without replacing clinical judgment.

What it does

AI Triage Assistant is a front-desk, conversation-based AI that estimates a patient’s Emergency Severity Index (ESI) level through structured dialogue alone. Using ESI-aligned reasoning, the assistant asks adaptive, context-aware questions to identify hidden risk signals, estimate urgency, and support waiting-list prioritization. The system does not diagnose or treat patients. It serves as a pre-triage decision-support tool, helping surface urgent cases earlier and reduce unnecessary delays for both patients and ER staff.

How we built it

The project contains three components: a python backend for replaying webhooks and storing data, a patient portal built on Next.js with Elevenlabs API, and a nurse portal built on Next.js.

Challenges we ran into

We explored integrating external APIs for video or voice-based vital sign detection, but access was not available within the project timeline. This led us to focus on a conversation only, front-desk triage approach.

Accomplishments that we're proud of

We spent a lot of time talking through ideas, refining our approach, and carefully designing how the AI should ask questions and listen to patients in a realistic ER setting. Getting the prompts right was important to us, because we wanted the AI to gather information that would actually be useful in real triage situations. That focus meant we had less time to build out every part of the system, but as a team we stayed aligned and worked efficiently to deliver a prototype that reflects what we originally set out to build.

What we learned

Through the hackathon, we learned how challenging it is to apply AI to real-world healthcare problems, where accuracy, safety, and context matter more than speed or flashy features. We realized that designing an AI system for healthcare requires careful thinking about how information is gathered, interpreted, and used, especially when decisions are made with limited data. We also learned the importance of scoping: deciding what to focus on and what to leave out. Working within constraints pushed us to make clearer, more thoughtful design choices. Most importantly, we learned how collaboration, iteration, and communication help turn complex ideas into a working prototype under tight time pressure.

What's next for AI Triage Assistant

Next, we aim to:

Integrate vital signs and real-time clinical data once available
Improve continuous re-evaluation as patient conditions change
Pilot the system in simulated or real clinical workflows to evaluate impact on wait times and triage accuracy Our long-term vision is an AI assistant that helps ensure no urgent patient is overlooked simply because urgency isn’t obvious.