Engineered a Siamese Neural Network with a BERT-pretrained backbone, increasing critical information extraction accuracy by 8% on 2,500 untrained invoices. Optimized data extraction by integrating an OCR model with a statistical classifier, achieving 86% accuracy in information categorization. Deployed the framework using FASTAPI, improving accessibility for 15,000+ users by optimizing performance, reducing latency, and enabling real-time data processing.