At NVIDIA, I co-built an XGBoost and random forest memory predictor across 1M+ EDA runs, achieving 77% exact-tier prediction with sub-second inference. I also co-owned root-cause analysis across 1.1M jobs, identifying policy-driven over-requesting and 56 PB per quarter of recoverable capacity.
Previously, I built production speech-to-speech booking agents at Travel Boutique Online using LangGraph, AWS Bedrock, GraphQL, and 11 live APIs, serving roughly 500 calls and 300 emails autonomously. My research work at Adobe and IIIT-Delhi spans in-context learning, RAG, LLM calibration, retrieval, fine-tuning, and evaluation, alongside competitive programming and ML competition experience.
