Edge and TinyML
Quantization, transfer learning, and shrinking models so they fit MCU / embedded budgets. I do this at Imagimob on DEEPCRAFT Studio.
work
What I help with
Most of my day job is the gap between a model that trains and a model that runs on real hardware: memory, latency, and a pipeline other people can operate.
Training pipelines
Scalable GPU training on AWS (EC2 G-series, Docker), tracking, and evaluation so experiments are repeatable — not a notebook on one machine.
Review and guidance
A short look at an existing model or pipeline: what will break on device, what to compress, and what not to bother training again.
Stack I use
Python, PyTorch, TensorFlow, ONNX, scikit-learn, Docker, AWS, GCP. C# and C++ when the product needs it (WPF, embedded tooling). CUDA at a working level. Git, Jenkins, TeamCity.