A SaaS platform I architected end to end for AI-powered jewellery image processing: a computer vision pipeline for tens of thousands of product photographs, serverless GPU inference across several regions, queueing that routes work to whichever lane has capacity, and the storage and billing around it. It serves 1,500+ users. The image work runs on ComfyUI, driven from the backend through BullMQ job queues with webhook delivery, over 6K-10K pixel images; how that is built is on the ComfyUI page.
Jewellery is an unforgiving subject - specular highlights, metal colour and stone clarity all break under a generic denoiser - so the networks are trained per material class rather than per image.
Anastasiia Butova - ML engineer, Belgrade, Serbia. Email black.design@me.com · LinkedIn · GitHub · Telegram · Hugging Face