Research
Research
I am currently pursuing thesis research under Prof. Shafkat Islam, Purdue University, at the intersection of Large Language Models (LLMs), Retrieval-Augmented Generation (RAG) and Differential Privacy. My work focuses on building AI systems that can reason over sensitive information while preserving confidentiality — ensuring that learning, retrieval, and output generation remain secure and controlled.
The broader goal is to enhance model intelligence without compromising data trust, enabling safe deployment of ML systems in domains where privacy is mandatory rather than optional. This direction holds strong relevance to healthcare, medical genetics, and privacy-aware AI infrastructure, where responsible reasoning can directly influence real-world adoption.
The research is currently under active development, with formal publication planned in the near future.
IEEE — ICACRS 2022 | Pudukkottai, India
A machine-learning-based framework designed to differentiate authentic and counterfeit Indian currency using image-feature extraction and classification techniques. Developed as an accessible solution for visually impaired individuals, the system identifies currency denomination and authenticity with high reliability.
The model extracts key visual markers using MATLAB, maps them to known patterns using ML classifiers, and achieves efficient recognition performance, addressing fraud-prevention and financial accessibility needs.
IEEE — CSITSS 2023 | Bangalore, India
A research-driven 4–2 compressor architecture for tree multipliers, optimized to reduce switching activity and power consumption. The design incorporates low-power techniques including AVLS, LECTOR, and Adiabatic Logic, achieving ~35.18% lower power usage compared to traditional implementations.
Validated through simulation & performance metrics, the proposed circuit supports scalable integration into digital multiplier systems.
Open-Access Research — Unpublished Work
Chest X-Ray Data — NIH Clinical Center (Kaggle)
PneumoGAN is a GAN-based medical imaging framework designed to enhance pneumonia detection using chest radiographs. The system trains a custom generator–discriminator architecture to create synthetic X-ray images, increasing dataset diversity and improving model robustness.
Achieving a 94.99% detection accuracy, the model demonstrates the clinical potential of synthetic augmentation in deep-learning-based diagnostics.