Marine Organism Detection

The problem

Identifying marine organisms under a microscope is slow, manual, expert work. The Smart India Hackathon problem statement asked for something a field lab could actually use: classify organisms from a live microscopy feed, in real time, on hardware cheap enough to deploy.

What we built

An end-to-end ML inference pipeline running entirely on a Raspberry Pi. Live video comes off the microscope, OpenCV preprocessing isolates organisms in each frame, and a trained classifier identifies them on the fly, with no cloud round-trip and no GPU.

The hard part was making “real time” true on an ARM-class CPU. We optimized the preprocessing and inference stages for the Pi’s memory and latency budget, trimming the pipeline until frames were classified as fast as the microscope produced them. This was an edge-deployment problem as much as an ML problem — exactly the intersection I like working in.

Result

We were named National Winner of the Smart India Hackathon 2025, Hardware Edition — the Government of India’s national innovation challenge. The Grand Finale ran December 8–12, 2025 at BHU Varanasi, with live judging on working hardware.