Completed AI Safety Fellowship. Red Teamed against unlearned models to reveal a concerning uptick in PII data leakage, measuring an increase from 8% to 20%.

Designed parallel routing and communication strategies for MoE generative models. Improved inference latency by 1.5-2.5x while achieving up to 98% GPU utilization.

Developed a GBDT-based ranking system to prioritize healthcare insurance A/R follow-up cases, reducing manual operational workload by 2.5 hours per day.

Built pothole detection systems using Mask R-CNN. Developed road asset segmentation models using U-Net architectures and implemented MLOps pipelines.

Improved NLP classification models for finance by 20% in f1-scores. Built hybrid transformer pipelines to extract graphs from corporate financial PDFs.

Merged Molecular Dynamics and Topological Data Analysis to create drug insights for Alzheimer's disease. Resulted in a published peer-reviewed journal paper.