I am a Data Scientist at HiLabs and an independent researcher associated with Vision and Learning Group, IIT Roorkee. Briefly, I have worked on Agentic Memory Management, Reinforcement Learning, Verifier Agents and OCR pipelines. I finished my graduation in 2026 at IIT Roorkee, where I was a part of the fantastic AI Research Group of the campus. I worked on Generative Engine Optimisation (GEO) while being an AI research intern at Infoedge.
My current research revolves around Agents, RL and LLMs, and my prior work on Position Bias in Mamba and Transformer models is currently under review for NeurIPS. I also have contributed to open source platforms like GNU Octave in Google Summer of Code. Off the clock I am a middle / long distance runner, a swimmer and a powerlifter.
Investigated position bias in long-context retrieval across Transformer, Mamba state-space, and hybrid architectures. Found that Transformers favor evidence near both context boundaries, while state-space models mainly favor evidence near the end. Controlled experiments showed that effective context use depends jointly on architecture, model capability, task, and prompt design.
Rubric Decoupled Advantage Normalisation GRPO is a rubric-based reinforcement learning repository for aligning LLMs with complex, open-domain instruction-following tasks. It improves performance on public instruction-following benchmarks.
Hugging FaceQwen3-4B-RDAN-GRPO
A scalable remote-sensing pipeline integrating EarthMind-4B, RemoteSAM, and a BERT classifier for unified satellite-image analysis. DRISHTI stands for Deep Remote-sensing Intelligence System for Holistic Task Integration. Awarded a silver medal at Inter IIT Tech Meet 14.0.
Developed an interview practice platform that generates diverse, topic-specific questions and uses NLP to analyse spoken responses, identify grammatical errors, and provide corrections after the interview.