B.Tech Information Technology · Delhi, India
ML Engineer & Aspiring SDE building reliable, production-ready systems at the intersection of software engineering and machine learning — clean Python, robust data pipelines, and reproducible model evaluation.
|
Deep learning pipeline for classifying ships from satellite/aerial imagery using EfficientNet-B0/B3, ResNet-50 & DenseNet-121 ensemble. Features open-set unknown vessel detection, Grad-CAM explainability, and confidence calibration. Trained on 22,248 multi-modal images (RGB + Grayscale + Contrast-enhanced).
|
Unified SegFormer-based multi-task framework performing classification, semantic segmentation, and change detection on satellite imagery in a single forward pass. Fine-tuned on EuroSAT with cross-domain transfer from ADE20K. Supports FP16 mixed-precision training on T4/A100 GPUs.
|
|
Python library bridging high-level Python with CUDA C++ for GPU-accelerated vector math. Achieved 3,100x speedup over pure Python (~0.8ms vs ~2.5s on 10M-element vectors) using custom NVCC kernels, Pybind11 bindings, and Tensor Cores on an NVIDIA RTX GPU.
|
Built a lightweight deep learning framework from scratch using CUDA C++ and Python, implementing tensor operations, automatic differentiation, neural network layers, optimizers, and GPU-accelerated training. Designed to mimic core PyTorch functionality while leveraging custom CUDA kernels for high-performance computation on NVIDIA GPUs.
|
| Domain | Technologies |
|---|---|
| Languages |      |
| ML / AI |       |
| Web |      |
| Databases |    |
| Cloud & DevOps |      |
