Inspiration
After a hospital stay, patients often receive several pages of discharge instructions while they are tired, stressed, or in pain. Important information about medications, appointments, tests, and warning signs can be difficult to understand.
I created ClairSortie to turn complex discharge documents into a clear and traceable action plan, without hiding uncertainty or inventing medical information.
What it does
ClairSortie allows a user to paste a synthetic or anonymized discharge document and generates:
- a simple next-step action plan;
- medication changes and follow-up tasks;
- the original source passage for each extracted item;
- questions to ask a healthcare professional when information is unclear;
- a short comprehension quiz;
- structured JSON data.
The application does not diagnose, prescribe, or replace a healthcare professional.
How I built it
I built a mobile-first web demonstration with HTML, CSS, and JavaScript and published it with GitHub Pages. The hosted version works directly in an Android browser and can run offline with synthetic demonstration data.
I also prepared a Python MVP using Gradio, pypdf, Pydantic structured data, automated tests, and an optional OpenAI API extraction path.
Challenges
The main challenge was making medical information easier to understand without turning simplification into unsafe medical advice. ClairSortie therefore prioritizes source citations, visible uncertainty, and professional confirmation.
Another challenge was developing and publishing the project using only an Android phone.
What I learned
I learned how to design a mobile web application, publish it with GitHub Pages, structure extracted information, and build safety constraints around a healthcare use case.
What's next
Next, I plan to use Codex to improve the project, strengthen automated evaluations, connect the hosted experience to GPT-5.6, add PDF highlighting and OCR, support more languages, and test the workflow with healthcare professionals.
ClairSortie is currently a hackathon prototype and is not a validated medical device.
Log in or sign up for Devpost to join the conversation.