Shelf-audit vision model for retail
Labeled 400K+ shelf images with bounding boxes and SKU-level tags, then trained a detection model to flag out-of-stock and misplaced items in real time.
96.4%
Detection accuracy
400K+
Images labeled
A sample of projects across the pipeline — some full end-to-end builds, others a single stage handed off to an in-house team.
Labeled 400K+ shelf images with bounding boxes and SKU-level tags, then trained a detection model to flag out-of-stock and misplaced items in real time.
Built a React Native app with an on-device document verification model, cutting manual review time for new account signups.
Annotated de-identified clinical notes for entity and urgency tagging, then fine-tuned a triage model to flag high-priority cases for review.
Replaced a legacy internal tool with a Next.js dashboard for dispatch and fleet tracking, cutting page load times and daily support tickets.
Ran response ranking and preference labeling at scale, plus a review dashboard for the research team to audit annotator decisions.
Trained a computer vision model on annotated field imagery and shipped it inside an offline-first Flutter app for use in low-connectivity areas.