Nilesh Kumar Parmar

I've built production systems for IPSERLAB clients DysputeAI and Feedzai, replacing a critical external LLM dependency with a fine-tuned Qwen model, cutting verification latency from about 900ms to 400ms, and building fraud, document-intelligence, and API credential services on AWS.

Previously, I delivered idempotent payment and retry systems at Paras Industries that eliminated duplicate charges and recovered $2K+ in daily revenue. I also fixed C++ memory leaks and race conditions in NVIDIA GPU driver code, while expanding automated test coverage across 10+ GPU architectures.