Inspiration

Through our friend's experiences working at doctors' offices, we were reminded how deadly colon cancer can be—and how a single miss by a radiologist can change a life. We believe technology should reduce that risk, not add to it. We named the project Semicolon because we don’t want to stop at colon cancer; our goal is to expand to many cancers.

What it does

Semicolon lets a user input clinical data about a colon tumor, upload a tumor scan, or use both together. The system predicts whether the tumor is malignant or benign, with the option to fuse modalities for higher confidence.

How we built it

We built the backend with Python and FastAPI, using TensorFlow to train and test our computer-vision model and to serve predictions to the frontend. The UI is built with Vite + React and Tailwind CSS for a fast, clean user experience.

Challenges we ran into

Time was our biggest constraint. Connecting the frontend to the backend reliably under hackathon pressure was tricky, and getting a model trained, tuned, and evaluated quickly required careful prioritization.

Accomplishments that we’re proud of

We shipped a smooth, responsive frontend and successfully ran our computer-vision model on tumor images end-to-end. Seeing the full flow—from input to prediction—working was a major milestone.

What we learned

We deepened our understanding of TensorFlow, improved our TypeScript/React skills, and learned how to stand up a lightweight FastAPI service to bridge ML and the web app.

What’s next for Semicolon

Next, we’ll implement the clinical-data ML component and fuse it with imaging for stronger predictions. We plan to benchmark our model against Gemini to compare accuracy and reliability. We’ll also complete the patient-to-doctor connection: matching users to nearby specialists in their specific cancer type and factoring in location, insurance, and other compatibility signals.