Kunal Singh

I've built production AI and data systems for Veersa Technologies, DRDO, and US healthcare workloads. At Veersa, I optimize ETL pipelines, SQL queries, and HealthEdge database schemas to improve data reliability and accessibility.

I developed Mapper (Megatron), an agentic schema-mapping platform that combines heuristics, FAISS semantic memory, target-schema similarity, and LLM reasoning to map source CSV columns to healthcare schemas. I also added automated profiling, human review workflows, tenant-specific MongoDB memory, rate limiting, caching, and a fallback mode for reliable operation.

At INMAS–DRDO, I built an intracranial hemorrhage detection and volume-estimation model from CT scans that achieved 81% mAP@50-95. I also developed a CMOS sensor-based nuclear radiation detection model reaching 98% Top-1 classification accuracy and deployed a VR-based CBRNe drone simulator on Microsoft HoloLens using Unity.

I'm focused on applied AI systems that move from models and data pipelines into useful production workflows, including LLM agents, computer vision, and real-time voice automation.