MedScribe

Inspiration

We wanted to build an app that helps patients actually understand what happened at their doctor's visit — because medical jargon shouldn't be a barrier to your own healthcare. Too many patients leave consultations confused, and we wanted to fix that.

What it does

MedScribe takes structured clinical data from a patient visit and transforms it into a warm, plain-language summary that any patient can read and understand and looks through NIH and FDA websites to suggest more medications for the doctor. It strips out medical jargon, explains medications in simple terms, and tells patients exactly what to watch for and what to do next.

How we built it

We built MedScribe using Python and FAST API for the backend, with OpenAI's GPT-4o-mini handling the AI summarization through a carefully engineered system prompt. The pipeline runs incoming clinical JSON through a validator, an LLM service, and a post-processing jargon filter before returning both a structured summary and a rendered HTML output.

Challenges we ran into

One large challenge we faced was the speech to text. We were able to process all the text but integration with muti-lingual capabilities and splitting the recording into doctor and patient was very challenging.

Accomplishments that we're proud of

We're proud of building a full end-to-end pipeline from raw clinical data to a patient-ready summary in a single API call, with a jargon blocklist safety layer on top of the LLM as a secondary guardrail.

What we learned

We learned how important it is to design around imperfect data — in a real clinical environment nothing comes in clean, and building a system that degrades gracefully instead of breaking entirely is just as important as the feature itself.