RAVIL PATEL

Currently developing an Agentic Framework which is especially designed for Chest Xray interpretation.

Forked Ultralytics, implemented custom modifications, and cut inference downtime by 60%, and published repository.

Developed and deployed AI-driven detection and heatmap generation modules, optimizing inference in production.

Trained and deployed custom object detection models with Distributed Data Parallel (DDP) using torchrun, optimizing efficiency and reducing time and resource costs.

Built and Deployed a high-resolution bone suppression model in, enhancing chest radiograph interpretation by reducing bone obstructions and improving soft tissue visibility.

Collected and annotated 5,000+ action-specific pet frames, enabling supervised model training.

Worked under Samsung's camera solutions team to translate business requirements into data-driven solutions.

Trained a lightweight Pet Action Recognition Engine using custom Efficient Net architecture, with handwritten loss function achieving 92% training accuracy.