Case study · 2025
Pepsi SKU Detection
Computer vision · Full-stack deployment
Segmentation model on 1,000+ hand-annotated images, 90% mAP@50, precision and recall, deployed as a full-stack app with live results.

mAP@50 · precision & recall
hand-annotated cooler images
Problem
Manual audits of retail shelves and beverage coolers are slow, error-prone, and leave inventory and planogram data stale for commercial teams.
Approach
Custom deep learning segmentation trained on over 1,000 hand-annotated multi-SKU cooler images, then shipped as HaidriX — a full-stack React and Flask application with live inference results.
Outcome
Exceeded 90% mAP@50 across precision and recall. Won Best Project Award at PepsiNorth (2025) and supports automated planogram and shelf analytics at Haidri Beverages.