I work on distributed systems security: Byzantine consensus, privacy in distributed learning, and the reliability of LLM safety monitors.
Distributed systems securityByzantine fault tolerance & consensusPrivacy in distributed and federated learningAI/LLM safety and multi-agent reliability
I'm pursuing a Ph.D. in Computer Science at the University of Kentucky. At the SDS Lab I work on security research: helping distributed consensus protocols adapt to changing network conditions with reinforcement learning, building zero-knowledge applications, and studying what gradients shared during distributed training leak about private data. On the side I look at why stacking multiple LLM safety monitors gives less protection than the independence math predicts.
Before the Ph.D. I spent over three years as a software engineer across several companies, building backend and real-time systems. That shaped how I think about what a system should do when conditions get messy and someone is actively trying to break it.
I earned my bachelor's degree in Electronics and Communication Engineering from the Institute of Engineering (Thapathali Campus), Tribhuvan University, Nepal.
Measured how much redundancy multi-monitor LLM safety systems actually provide. Across 15 monitors from three providers, monitors failed together far more often than independence assumptions predict — committees of five behaved like roughly two independent monitors.
Showed that the policy gradients shared in distributed reinforcement learning leak temporal structure across training steps, and built an attack that reconstructs full private observation–action sequences from those gradients alone.
Designed a switching layer that lets a running BFT system move between synchronous leader-based and asynchronous leaderless protocols as network synchrony shifts, with replicas deciding in real time from observed metrics.
