Sagar Sahayak - Pytorch OpenCV Scikit ONNX OpenVINO NumPy Pandas

Architected a physics-informed synthetic data engine pipeline, fusing MiDaS depth estimation with Jaffe-McGlamery light models to procedurally generate photorealistic training datasets, backed by a Grounding DINO auto-annotation pipeline for zero-shot labeling. Engineered a dual-stream neural architecture inspired by Pix2Pix GANs for real-time image restoration (turbidity removal) and YOLOv11 for maritime threat detection, achieving high-fidelity visibility in degraded underwater environments. Optimized inference for edge deployment via Intel OpenVINO & NNCF, implementing INT8/FP16 posttraining quantization to achieve a 75% model size reduction and 4x throughput acceleration on heterogeneous CPUs/GPUs. Deployed a production-grade Streamlit interface with role-based security, offering real-time model benchmarking (PSNR/SSIM/UIQM) and vendor-agnostic interoperability across TensorRT and CoreML ecosystems. Large initiative for DRDO of India With custom developer toolkits, and specialized AI models