I build things that learn. Not because it's trending, but because there's something genuinely satisfying about watching a system improve at a problem you couldn't just hardcode your way out of.
My work spans the ML stack: from classical models and deep learning to reinforcement learning; applied across healthcare, environmental forecasting, and autonomous systems. But more than any specific domain, I care about whether something actually works outside the notebook.
I'm a try-athlete; not fast, not consistent, but always attempting something new. That approach carries into how I work: I'd rather try a method, fail fast, and learn than wait for perfect conditions.
Syracuse University
Cognifai
Omdena
Zummit Infolabs
Researched transformer-based models to classify Individual Case Safety Reports into Valid, Potential, and Invalid categories for pharmacovigilance compliance.
Fine-tuned and benchmarked biomedical NER models on PubMed datasets to extract patients, drugs, and adverse events from medical literature using SciBERT and SciSpacy.
Built scalable pipelines for CAQI-based air quality forecasting using pollutant and weather time-series from 300+ stations, integrating MSTL decomposition and spatial harmonization.
Developed an audio-to-video synthesis pipeline using conditional GANs and 3D Morphable Models to generate photorealistic talking-face animations from raw speech. Achieved 25% improvement in lip-sync accuracy and raised SSIM from 0.76 to 0.91.
Trained a PPO reinforcement learning agent in a Gym-wrapped CARLA simulator, achieving 95%+ autonomy with intervention rates below 2/km across complex urban driving scenarios.
Engineered predictive models from 3D spinal movement data to provide personalized and gamified rehabilitation plans for back pain relief using signal processing and motion analytics.
Syracuse University
Gujarat Technological University