Research
Areas of inquiry, open questions, and ongoing investigations.
Cybersecurity
Focus on securing IoT gateway architectures and connected autonomous vehicle sensor streams against false data injection, botnets, and eavesdropping.
Explainable AI
Developing interpretability methods to make ML models trustworthy in high-stakes applications. Work spans SHAP, LIME, and attention-based explanations for adversarial and IoT threat detection scenarios.
Machine Unlearning
Investigating algorithms that enable deep learning networks to selectively forget training samples on demand, complying with right-to-be-forgotten privacy mandates without full model retraining.
NLP & LLMs
Engineering Retrieval-Augmented Generation (RAG) and conversational agents, with focus on securing natural language to SQL translators and voice-assisted workflows.
Quantum Security
Applying quantum cryptographic primitives like Continuous-Variable Quantum Key Distribution (CV-QKD) to secure resource-constrained IoT nodes and vehicular networks.
Research Journey
My research began in early 2025 under the supervision of Prof. Sudeep Tanwar and Prof. Rajesh Gupta at Sudeep Tanwar's Research Group, initially exploring how explainability tools like LIME could make intrusion detection systems legible. What started as a question about smart-home security quickly expanded: if you could explain a prediction, could you also diagnose a poisoned training set? That forensic inversion became a thread connecting papers across cybersecurity, autonomous vehicles, and healthcare AI. At IEEE ICSC in Tampa, presenting in person to a room of practitioners taught me something no peer review ever had — that correctness and deployability are different standards. I came back asking the deployment engineer's questions about every paper I write. This led to exploring quantum security primitives like Continuous-Variable Quantum Key Distribution (CV-QKD) to secure resource-constrained edge nodes, moving beyond computational assumptions to physical-law security guarantees. The unlearning verification problem and quantum-secured telemetry streams are where that thinking has landed for now. Seven papers in eighteen months, and the questions are still getting harder.