RetiNova
RetiNova is a healthcare-focused computer vision application that detects eye conditions from retinal images, designed and shipped end-to-end within a 36-hour hackathon.
What it is
RetiNova is an AI-powered healthcare application that enables users to upload retinal images and receive real-time predictions for potential eye conditions like cataracts or uveitis. The system ties a trained computer vision model (OpenCV/PyTorch) to a production‑ready web frontend and a lightweight API layer, focusing on turning raw ML outputs into a usable screening workflow under tight hackathon constraints. The goal was practical accessibility: fast feedback, simple UX, and a bridge from detection to care.
My Role
I led backend and ML integration. I implemented the inference API that connects PyTorch models to the UI, built the geolocation and clinic search feature (Nominatim/Overpass), and containerized the frontend, backend, and model services for smoother deployment. I also coordinated with teammates on UI polish and motion to keep the interface clear and responsive while we iterated quickly.
Interesting Constraints
- 01Training data was limited and fragmented so we had to source, clean, and balance datasets quickly to avoid biased predictions.
- 02Real‑time feedback meant minimizing inference latency and API overhead so results felt instantaneous.
What I Learned
- Practical ML in products lives at the interface: clean APIs, fast inference, and unambiguous UI matter as much as model metrics.
- Containerizing the whole pipeline early (model + API + UI) reduces integration risk and accelerates iteration.
- Planning and tight task delegation are critical in hackathons. Having clear boundaries and priorities keep shipping speed high.