I’m Priyanshu Karmakar, a first-year PhD student in CSE  @IIT Bhubaneswar, where I work primarily with Large Language Models. My research sits at the intersection of rigor and real-world utility: building reliable LLM systems with strong evaluation, practical guardrails, and retrieval/tool use that actually helps users get things done. I care about clean baselines, reproducible experiments, and shipping models that balance quality, latency, and cost. Recently, I’ve been exploring rubric-driven judge models, lightweight inference stacks (batching, caching, quantization), and retrieval pipelines that make LLMs more grounded and less prone to drift.