I built a local AI system to detect duplicate product listings before they go live. The problem was simple but expensive: The same product was being uploaded by different sellers with different titles, descriptions, and keywords. Traditional text search couldn’t reliably catch these duplicates, so moderators had to review hundreds of listings manually. My solution was to compare the product images instead of the text. How it works: → A seller uploads a product image → CLIP converts the image into a 512-dimensional embedding → The vector is stored and searched in Qdrant → Similar product images are found in milliseconds → If the similarity score is above 0.85, the listing is flagged for review before publishing What makes this system useful: ✔ Automatic duplicate product detection ✔ Less manual moderation ✔ Faster listing approvals ✔ Fully local AI processing ✔ No per-request AI API cost ✔ Fast vector similarity search Tech Stack: FastAPI · Python · CLIP ViT-B/32 · Qdrant · Pillow Ideal for: • E-commerce platforms • Online marketplaces • Classified applications • Large product catalogs • Content moderation systems This project shows how computer vision and vector search can solve a real marketplace problem that keyword-based search cannot. If you're building a marketplace or e-commerce platform and need AI-powered image search, duplicate detection, or product matching, feel free to message me.

abdullah masood

AI & Full-Stack Engineer | Building Production AI SaaS, APIs