ID Scanner Application

Problem

Real-world ID scanning involves poor lighting, motion blur, damaged documents, and inconsistent layouts. Cloud-based OCR introduces latency and privacy concerns, especially in regulated environments.

Solution

Built a fully on-device processing pipeline that detects documents, guides users during capture, and extracts structured data using optimized computer vision and OCR techniques.

Architecture & Approach

React Native application using react-native-vision-camera with custom frame processors. Google ML Kit is used for OCR and face detection. All processing runs locally without requiring cloud connectivity.

Key Tradeoffs

• On-device processing limits model flexibility but eliminates latency and privacy risks
• Supporting many document types requires continuous updates as formats evolve
• User guidance during capture is more effective than post-processing low-quality images

What I Learned

• Edge ML requires careful optimization to meet performance constraints
• Real-world data is significantly messier than sample datasets
• Immediate visual feedback during capture dramatically improves accuracy