Published Paper Details:

This paper presents an intelligent Optical Character Recognition (OCR) system for automated bill data extraction using Qwen VL2, a state-of-the-art vision-language model (VLM). Unlike traditional OCR pipelines that rely on Tesseract or rule-based pattern matching, the proposed system leverages the multimodal capabilities of Qwen VL2 to process bill images and directly return structured key-value pairs corresponding to fields such as consumer number, billing amount, due date, and service period. The system integrates a Node.js backend with a MongoDB database for scalable storage and retrieval. Experimental evaluation on 200 diverse real-world electricity and utility bills demonstrates superior extraction accuracy compared to conventional OCR approaches, achieving 94.6% overall field-level accuracy versus 73.0% for Tesseract. The architecture is lightweight, extensible, and deployable in cloud or on-premise environments.