video1.0<iframe src="https://www.loom.com/embed/78bb9817a702454cae2de36e87a441d4" frameborder="0" width="1218" height="913" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>9131218Loomhttps://www.loom.com9131218https://cdn.loom.com/sessions/thumbnails/78bb9817a702454cae2de36e87a441d4-4f24d35105dfee24.gif377.197I walked through the AI Morph Identification Pipeline for Reptidex from the web client down to inference. The workflow supports up to three image angles, compresses each to 1200 by 1200 at 0.8 JPEG quality, and sends them in one SageMaker call via Gecko0, Gecko1, Gecko2. On the server I hash the image buffer with SHA 256, check a prior results table for cache hits to cut costs, and run images in parallel with promise.all. I also version the scoring with AgreementMax v1 and apply up to a 5 percent boost when traits agree. There was no action requested from viewers.