AI Stroke Outcome Prediction Platform

Problem

Clinicians need a straightforward way to submit MRI scans and patient assessment data to an AI model that predicts stroke outcomes, then follow each job through to its results.

My role

• Built the Next.js and TypeScript app that lets clinicians upload MRI scans (NIfTI, up to 500 MB), enter clinical assessments such as NIHSS and Fugl-Meyer scores, and submit prediction jobs.
• Implemented the API routes for creating, listing and tracking jobs, backed by Prisma on AWS RDS MySQL and AWS S3 file storage.
• Implemented the pages for new predictions, job history and result viewing, including a 3D brain viewer (Niivue) with an adjustable lesion overlay.
• Added client-side PDF reports and segmentation-mask downloads for completed jobs.
• Added mock and simulation modes alongside production (in-memory database, simulated processing, local files), so the app can be developed and demoed without the database or AI service.

Challenges

• Tracking each job through an eight-stage pipeline, from upload through preprocessing, segmentation and classification to a result or an error.
• Rendering MRI volumes and lesion masks in the browser with WebGL across axial, coronal, sagittal and 3D views.
• Validating large MRI uploads and every clinical score against its accepted range before a job is submitted.

Results

• Sole developer: wrote all 81 commits between December 2025 and March 2026.
• Wrote 157 unit tests (Jest + React Testing Library) plus Playwright end-to-end tests.