Biography

Chandrav Rajbangshi is an AI Research Intern at RAAPID Inc. and a B.Tech student in Computer Science and Engineering at Birla Institute of Technology, Mesra. His research focuses on geometry-aware parameter-efficient fine-tuning (PEFT) for large language models. At RAAPID, he co-authored GRIT, a curvature-aware adaptation framework that combines Fisher-guided reprojection, K-FAC–based natural gradient preconditioning, and dynamic rank adaptation to stabilize low-rank updates. His work introduces a principled scaling law to model catastrophic forgetting through effective rank and Fisher alignment, achieving significant efficiency–quality trade-offs across instruction tuning and reasoning benchmarks. Beyond research, Chandrav builds end-to-end ML systems spanning vision, NLP, and retrieval-augmented generation, with an emphasis on translating theoretical insights into robust, scalable implementations