Veerpal Singh

• Cloud & AI Architecture: Architected a scalable and cost-effective Azure cloud infrastructure to process over 100,000 documents daily, utilizing Azure Kubernetes Service, Azure Functions, PromptFlow, and Azure OpenAI.
• Document Classification Model: Developed a hierarchical, text-based classification system by fine-tuning transformer models like DistilRoBERTa.
• Advanced Data Extraction: Engineered a sophisticated data extraction pipeline for large, complex, structured, and unstructured documents using GPT-4o, enhanced by an advanced Retrieval-Augmented Generation (RAG) framework.
• Computer Vision & OCR: Implemented a novel method for handwritten signature detection by correlating OCR text coordinates with computer vision analysis via OpenCV, increasing detection accuracy and speed.
• RAG-Based Document Querying System: Engineered an end-to-end Retrieval-Augmented Generation (RAG) system for natural language querying of large-scale PDF documents. Orchestrated a Llama 2 LLM with LangChain, and utilized vector databases (FAISS, Pinecone) for high-speed, accurate semantic retrieval.
• Compiled, cleaned and manipulated data for proper handling.
• Analyzed large datasets to identify trends and patterns in customer behaviors.
• AI Agent and Copilot Development: Designed and deployed custom AI agents like Sales FAQ , Meeting Brief Generator, meeting scheduler etc and enterprise copilots using Microsoft Copilot Studio to automate business processes and enhance user interaction with internal systems.
• API Test Automation: Engineered a robust, Python-based test automation suite within the Robot Framework for end-to-end API validation. Authored custom Python libraries to create reusable API functions and perform complex data validation on response payloads, significantly improving test reliability and efficiency.

Image Pre-processing for developing Custom OCR:

• Architected and implemented a multi-stage image enhancement pipeline to prepare documents for high-accuracy text recognition.
• Rotation Correction: Developed and trained a custom Convolutional Neural Network (CNN) to automatically detect document skew, using a specialized dataset and extensive data augmentation for robustness.
• Image Denoising: Implemented image processing techniques to remove noise, speckles, and other artifacts, improving the clarity and legibility of text on degraded documents.