Computer vision to catalogue Geev

Geev is a mobile app dedicated to peer-to-peer donation of items and furniture: the logic of classifieds, fully tuned to giving. Hundreds of thousands of items are posted every year, often mis-categorized by users at publication time. A direct drag on discovery, receiver trust, and the model's ability to scale.

Categorize every item precisely from its photo alone. Not just "sofa" or "cabinet", but at the granularity that matters to the user: 2-seater sofa or corner sofa, small cabinet or large cabinet, coffee table or dining table. At industrial volume, without systematic human oversight.

Computer vision pipeline built on the OpenAI API. Every photo posted in Geev is analyzed to extract the main object, then a fine classification assigns sub-category, size, and use. Human evaluation loop on ambiguous cases, distributed batch processing to handle the continuous publication load.

Delivered at a time when computer vision via OpenAI wasn't yet a mainstream reflex. The available models demanded careful prompt engineering, a solid taxonomy, and fallback logic to reach production-grade precision at scale.

Over 500,000 items finely re-categorized, industrialized tool for continuous processing of new listings. Direct impact on in-app discovery, a concrete contribution to scaling peer-to-peer donation.