Ongoing Research Projects

Ph.D. Research, MasonNLP Research Group, George Mason University [2024 - Current]

• Agentic AI for Medical Error Detection and Correction. Five specialist reasoning agents orchestrated across a LangGraph DAG with sequential context dependencies, adjudicated by a coordinator constrained to cite only agent-flagged evidence, and grounded by retrieval-augmented generation over a FAISS vector index built from FDA drug labels ingested through the OpenFDA and RxNorm APIs. (Python, LangGraph, LangChain, FAISS, LLM APIs, Pydantic) — Code
• LLM-Driven ETL to the OMOP Common Data Model. Converting heterogeneous electronic health record databases into a standardized common data model, benchmarking seven large language models across MIMIC-IV, OpenMRS, and Synthea on executability, content fidelity, and referential integrity. (Python, SQL, LLM APIs, OHDSI Athena)
• Cascade Prompting for Automated Document Classification. A cascade prompting framework for classifying documents by type and content, with a comparative evaluation of open-weight and proprietary large language models. (Python, LLM APIs, prompt engineering) — Code
• Automated Identification of Incidentalomas from Radiology Reports. Multi-anatomy evaluation of generative LLM-based and supervised approaches against a human inter-annotator agreement ceiling. (Python, PyTorch, Hugging Face Transformers, LoRA)
• Benchmarking and Evaluation of Large Language Models. Controlled ablation design, paired significance testing, and confidence interval estimation for open-weight versus proprietary models under zero-shot, few-shot, and context-aware prompting. (Python, PyTorch, HPC GPU clusters)

Earlier Research Projects

Explainable AI, Computer Vision, and NLP [2019 - 2023]

• Explainable Cost-Sensitive Deep Neural Networks for Brain Tumor Detection. Cost-sensitive learning under severe class imbalance in brain MRI, with Grad-CAM interpretability. Published in Multimedia Tools and Applications (Springer), 2025. (TensorFlow, Keras, Scikit-learn, Grad-CAM)
• Bengali License Plate Recognition. License plate restoration and recognition combining a convolutional network with GFP-GAN image enhancement. Published at ICCIT 2023. (Python, PyTorch, OpenCV)
• Interpretable Movie Review Analysis. Sentiment classification with machine learning and transformer models, explained with LIME and SHAP. Published at IEEE CSDE 2022. (Python, Hugging Face Transformers, Scikit-learn, LIME, SHAP)
• Bengali News Headline Categorization. Empirical comparison of machine learning techniques for multi-class categorization of Bengali news headlines. Published at ICCIT 2022. (Python, Scikit-learn, NLTK)
• An Empirical Study on How Developers Discussed Pandas Topics. Topic modeling and analysis of developer discussion text. Published at MIET 2022. (Python, LDA topic modeling, Scikit-learn)
• Interpretable Retinal Disease Classification from OCT Images. Deep neural network classification of retinal disease with explainable AI. Published at ICECIT 2021. (Python, TensorFlow, Keras, Grad-CAM)
• A Comprehensive Survey of COVID-19 Detection Using Medical Images. Survey of deep learning approaches to COVID-19 detection from X-ray and CT imaging. Published in SN Computer Science (Springer), 2021.
• Early Depression Detection from Social Media. B.Sc. thesis project applying deep learning to detect early signs of depression in user-generated text. Published at IEEE TENSYMP 2020. (Python, TensorFlow, Keras) — Thesis

Undergraduate Course Projects [2015 - 2019]

• Benchmarking Different Versions of BERT (Python, TensorFlow).
• Sentiment Classification (Python, Scikit-learn, TensorFlow).
• Image Caption Generator (Python, TensorFlow).
• A website to draw 2D and 3D objects, with transformation algorithms implemented to visualize data (WebGL, C++).
• Food Finder Website (PHP, MySQL).
• Bluetooth Controlled Matrix Signboard (Arduino).
• IELTS Exam (Java Servlet, Eclipse).
• Online Computer Shop (C#, MySQL).
• Magnificent Cox’s Bazar (Android Studio).
• Digital Logic Design Simulator (Java, NetBeans IDE).
• Mario Hopper (C, iGraphics).