At Neodocs Healthcare Pvt. Ltd. (YC S21), I design and deploy real-time computer vision pipelines using YOLOv5, OpenCV, and TFLite, including on-device inference with sub-200ms latency.
I developed image preprocessing workflows that reduced image rejection rates by 65% across multiple devices. I also led model training and optimization for detection and classification, achieving up to 96% accuracy and 99% sensitivity in production.
For document understanding, I developed NLP and LLM-powered pipelines with LangChain and integrated LLM-based reasoning with ML predictions to automate decision logic, validation, and reporting workflows.
In my Patient Test Companion project, I built an agentic assistant for urinalysis and kidney function results, using LLM function-calling over deterministic clinical tools. I designed hybrid retrieval with Qdrant, RRF fusion, and cross-encoder reranking, and integrated a custom Google Document AI extractor for uploaded lab reports.
