Lead the Computer Vision team, overseeing development and implementation of advanced vision-based solutions using state-of-the-art models including Vision Transformers (ViT) for PSE, MediaPipe, YOLO, and HMR4D for 3D mesh generation.

Spearhead model optimization initiatives using TorchCompile and deploy production-ready models with optimized formats (TFLite, ONNX), significantly improving inference performance for on-device applications.

Design and implement end-to-end computer vision pipelines, integrating motion data analysis using OpenSim, ensuring robust performance from data ingestion through model deployment.

Drive pipeline optimization efforts, focusing on efficiency improvements in model training, inference, and deployment workflows while maintaining high accuracy and real-world applicability.

Model Training & Maintenance: Expertise in building and refining high-performance SDXL models for computer vision applications. Leverage control net techniques for enhanced model optimization.

Data Cleaning & Preprocessing: Utilize OpenCV and other libraries to meticulously curate real-world data (photos, videos, simulations) for efficient AI training. Ensure high-quality data through thorough cleaning and image processing techniques.

Prompt Engineering: Craft clear and concise prompts for optimal AI model guidance (e.g., SDXLs) to extract valuable insights from visual data. Develop specific prompts for enhanced image generation within datasets for targeted training.

Computer Vision Research: Stay current on cutting-edge advancements in computer vision, with a focus on identifying and implementing the most effective solutions for specific tasks.

Led the development of a Generative AI-powered enterprise solution (fine-tuned Llama2) for extracting insights from any data format. Built robust agents that enabled users to converse with their data and unlock insights for forecasting and analysis.

Developed a robust and lightweight Pneumonia detection model for Swasth AI, achieving 95% accuracy, using Convolutional Neural Networks (CNNs) and image processing techniques.

Led seamless deployment of DOD on GCP using CI/CD pipelines, virtual machines, and load balancers for scalability. Implemented Docker containers for efficient packaging and deployment, ensuring reliability and scalability.