MediCura

Medical paperwork is written for the chart, not for the person holding it. At Uncommon Hacks I built MediCura so a lab PDF or a doctor’s note comes back in plain language.

The problem

A lab printout assumes you already know which numbers matter and what the abbreviations mean. The person holding it is often reading it in a hallway, on a phone, without the chart in front of them. The frustration is not a missing feature. It is a document that will not explain itself.

We wanted a place to drop a PDF or a photo and get a glossary, the lab values, and a chat that stays educational rather than diagnostic. It was worth building because the alternative is retyping the form into a search box and hoping the first link is about your result and not a different one.

How I built it

The app is Expo, with Expo Router. I built the pixel-art home, the analyze screen, the chat, and the emergency screen. The emergency screen speaks, through Expo Speech, because an emergency screen that you have to read carefully is a bad emergency screen.

Analyze sends the file to OCR.space. A photo of a page is not text until something reads it. Then gpt-4-turbo writes the plain-language pass: the glossary and the values, in words a person can repeat. The OCR step is why a phone photo works at all. The model step is why the result is not just a second copy of the jargon.

The chat is an in-app assistant called Dr. Careo. It stays on the educational side of the line. The screen says it is not medical advice. That sentence is part of the product, not a footnote I added later.

Challenges and what I learned

The hard part was the boundary. A fluent answer about a lab value sounds like a diagnosis even when you asked for a definition. Keeping Dr. Careo educational, and saying so on the screen, is the constraint. OCR.space can misread a digit, which is another reason the pass is an explanation of the page and not a clinical decision.