Pipeline
Images land in S3, trigger Lambda preprocessing, and move through a PyTorch segmentation model. A FastAPI service exposes inference plus explainability overlays consumed by QA operators.
MLOps
Training pipelines log metrics to ClearML, and a drift monitor alerts when fat distribution changes. Rollouts happen via blue/green Lambda aliases to avoid downtime.
Business outcome
Buyers receive standardized reports in both Spanish and English, cutting manual review time by several hours per batch and aligning pricing decisions across plants.