FisherAI — Automated Shelf Card Generation

Overview

FisherAI processes scanned store ads and shelf imagery to extract product information and automatically generate shelf cards for in-store displays. The pipeline combines object detection, OCR, and layout-aware document understanding to robustly parse heterogeneous ad/layout formats and produce structured product metadata for downstream card rendering.

How it works

• Detect product regions and visual elements using YOLO-based object detection.
• Extract text from images and ads with PaddleOCR for multi-language OCR and robust recognition.
• Apply LayoutLMv3 to model document/layout context and resolve entity relationships (price, product name, brand).
• Post-process and normalize extracted fields, then generate shelf-card assets (images + metadata) automatically.

My role & tech

Work project — responsible for end-to-end pipeline design, model integration, and productionization.