Parth Patel

Architected enterprise VLM system combining Azure Document Intelligence with GPT-4 Vision, achieving 95%+ accuracy in legal document processing. Pioneered image-text reconciliation approach bridging traditional OCR limitations with AI contextual understanding. Built cloud-native architecture on Azure Container Apps with zero-shot learning for new document types without retraining. Architected comprehensive GenAI platform processing 1000+ legal documents with 97.56% accuracy. Built production-scale LLM infrastructure on Kubernetes achieving 68-88% cost reduction. Engineered 14 specialized transformer models (7 NER + 7 REL) with 95-98% precision. Designed intelligent document processing system with OCR integration and 20+ category classification. Achieved 60%+ effort reduction in contract analysis through innovative prompt chaining architecture. Led benchmarking study resulting in Instructor-XL model selection with 12.8% improvement over baseline. Developed multi-modal classification incorporating visual and textual document attributes. Built GPT-4 VLM RAG pipeline achieving 90.3% weighted accuracy and 94.0% completeness. Implemented Meta's Segment-Anything Model resulting in 20% increase in text completeness. Pioneered segment reduction algorithm scaling from 100+ to <30 segments using MECE principle.