Full-Stack ML Engineer | GPU Computing | Production AI Systems
- Building end-to-end ML pipelines — from data to production serving
- GPU kernel optimization on AMD ROCm and NVIDIA CUDA
- Model architectures — Vision Transformers, GPT, Diffusion models
- MLOps — experiment tracking, model registry, automated retraining
- ML/DL: PyTorch, ONNX Runtime, HuggingFace Transformers
- GPU: ROCm/HIP, CUDA, Triton
- Serving: FastAPI, TorchServe, ONNX Runtime
- Infra: Docker, Kubernetes, Prometheus
- Languages: Python, C++, Bash
| Repo | Description |
|---|---|
| rocm-kernel-lab | GPU kernel optimization (matmul, FlashAttention, LayerNorm) |
| pytorch-model-zoo | ViT, GPT-2, DDPM implementations |
| fastapi-ml-serving | Production ML serving with dynamic batching |
| cuda-benchmarks | GPU memory & compute benchmarking |
| mlops-pipeline | End-to-end training & deployment pipeline |
- GitHub: @soncel15