At the University of Manchester, I completed a postgraduate research project on event-based sign-language recognition using spiking neural networks and event-camera data. I compared two input conditions with an identical network to measure the effect of discarding non-hand data, and evaluated performance with subject-independent splits audited for leakage.
At Vellore Institute of Technology, I built AGFC-Net to classify faults in electrical power systems, reaching 99.52% accuracy across 12,000 voltage and current instances. I checked its performance with cross-validation and per-class measures, then wrote up the work for a paper published in Scientific Reports.
In a remote internship with VIT Chennai and Nagasaki University, I built DCSIN to classify six tea leaf diseases from field photographs. The model exceeded the previous best published accuracy by 4%, with much of the improvement coming from preprocessing and augmentation; the work was released as a preprint.
My other research includes safe reinforcement learning for autonomous-vehicle trajectory planning, published as a review in E-Prime, and a lip-reading web app with live inference that reached 97% sentence-level accuracy. I also built Zero OS, a multi-agent AI assistant with a FastAPI backend and React and Tauri desktop app.
