Document AI · Large Language Models · Multimodal AI · Robust Machine Learning
I am a machine learning researcher and engineer with a Ph.D. in Computer Science from the University of Texas at Arlington. My work focuses on large language models, Document AI, structured information extraction, multimodal learning, and evaluation of robust AI systems under realistic data-quality constraints.
My research combines academic investigation with industry experience building production-scale machine learning systems. I previously worked as a Senior Machine Learning Engineer at Iron Mountain, where I developed Document AI pipelines involving document enhancement, OCR, transformer-based information extraction, table detection, and LLM-powered question answering.
I am currently conducting independent research on LLM-based key–value extraction from forms and receipts, with an emphasis on reproducible evaluation, OCR robustness, prompt sensitivity, and failure analysis. My broader goal is to build document understanding systems that remain reliable when deployed on noisy, visually complex, and imperfect real-world data.
Current work in robust document understanding, structured extraction, and evaluation of large language models.
A reproducible benchmark of six open-source instruction-tuned LLMs for key–value extraction across FUNSD, CORD, and SROIE under clean-text and OCR-degraded conditions. The study compares PaddleOCR, EasyOCR, and Tesseract and analyzes hallucination, numeric corruption, prompt sensitivity, and key–value misalignment.
A comprehensive survey and interpretive review of LLM-based and traditional approaches for key information extraction, named entity recognition, document classification, and document question answering, with emphasis on architecture, modality fusion, prompt sensitivity, and generalization.
An unpaired image-to-image translation framework for single-image dehazing using global-local discrimination, perceptual supervision, color consistency, residual learning, and skip connections.
A benchmark and dataset for evaluating single-image dehazing methods under non-uniform, realistically colored sunlight haze using full-reference and no-reference image-quality metrics.
Selected publications. See Google Scholar for the complete and most current list.
Research on LLM-based Document AI, robust key–value extraction, OCR sensitivity, reproducible evaluation, and hybrid document understanding systems.
Developed production Document AI systems involving document restoration, OCR, transformer-based information extraction, table detection, RAG, and model evaluation.
Worked on face and emotion recognition, super-resolution, object tracking, and model benchmarking.
Conducted research in deep learning, computer vision, image dehazing, and automated dataset generation.
I have several years of university teaching assistant (TA/GTA) experience across computer science courses, including machine learning, algorithms and data structures, parallel processing, computer architecture, software engineering, and computer systems. My teaching interests include machine learning, artificial intelligence, computer vision, data structures, and applied programming.