Industrial Projects
Commercial deep learning and computer vision solutions deployed in production environments, serving real-world business needs at scale.
Commercial Projects
National Postal OCR System
Developed a transformer-based OCR system for processing millions of postal scans daily across the national postal network. Implemented real-time inference pipeline with sub-second latency requirements and rigorous error handling for production deployment.
Impact: Processing 1M+ scans daily with <0.5% error rate
Object Detection and shrinkage detection for Retail Analytics
Built end-to-end computer vision pipeline for objecte detection and shrinkage detection in retail stores. Leveraged object detection models optimized for edge deployment and embedding based matching techniques for object identification between 50k+ SKUs.
Impact: Deployed across 5 stores in 100+ checkouts, 30% improvement in inventory management
Luggage tag recognition, reading and matching system for airports
Modernized the luggage tag recognition and reading systems used in major airports processing thousands of luggage items per day. Developed robust OCR and object detection models to handle diverse tag designs and challenging imaging conditions.
Proof of Sorting in warehouse logistics
Developed a computer vision system to verify correct sorting of packages in warehouse logistics. Implemented real-time image capture and analysis pipeline using edge devices and optimized deep learning models for object detection, tracking and action recognition.
Order Fulfillment Validation System
Developed a template based object detecion system utilizes ideal images of skus to verify correct order fulfillment and counting. Embedding based matching techniques are used to identify products from a large catalog
Note: Project details have been generalized to respect client confidentiality agreements.