Surya User

I’m an AI/ML intern building synthetic dataset pipelines for deepfake research, integrating open-source models like InsightFace and Qwen with automated labeling and metadata annotation.

I also work on OCR-based and document-level fake data generation for multimodal synthetic dataset creation, and I’ve engineered preprocessing and quality scoring to produce reproducible train/val/test splits with comprehensive metadata logging.

On the side, I developed OA² for organ-action detection using PyTorch, YOLO-style detection, and compact affordance modeling, reaching 95.9% organ prediction and 80.2% action prediction and deploying the end-to-end system on Hugging Face Spaces. I’ve also built Layer Insight, an automated PyTorch model profiling framework that captures CPU/GPU execution traces, operator stats, CUDA kernel activity, memory usage, and generates actionable optimization reports—along with strong competitive coding achievements (CodeChef 3 Star, Codeforces Pupil, and CodeClash runner-up).