Shubham Chauhan

Built an LLM powered patient chatbot for collecting symptoms and diagnostic context, optimizing rule-based routing to cut token usage by 70%. Developed a RAG-based Help Section chatbot for automating responses and classifying issues. Trained and deployed MaskRCNN and YOLOv11 models for skin condition detection. Built an EfficientNet-based Melasma severity classifier. Improved model training efficiency with gradient checkpointing and mixed-precision. Developed India's first segmentation-based baldness detection model. Built a clinical decision-support model with 98% accuracy. Implemented auto-scalable deployments using FastAPI, Ray Serve, and TensorFlow Serving on Kubernetes, reducing cloud costs by 63%. Developed internal tools for monitoring model performance and annotation frameworks.

Built face-tracking and skin-analysis pipelines using SOTA models (MediaPipe, 3DDFA) for automated reporting of skin condition diagnosis. Designed lesion-overlap and severity-scoring metrics. Built tools to aggregate multi-annotator labels using IoU, Cohen-Kappa, Fliess-Kappa, and other statistical methods.

Built an end-to-end Number Plate Recognition system using YOLOv5 and pytesseract with 93% accuracy. Evaluated camera types and setups to reduce distortion and improve model performance. Developed a Flask application to deploy the NPR system with a web interface.