Founding Edge ML Engineer

The radar detects a moving target. Is it a Mavic, a Shahed, a bird, or ground clutter? You have milliseconds to decide. You have no cloud connection and no heavy GPU. You are running inference on a tiny edge processor inside a vibrating seeker. You must solve this classification problem under severe memory bottlenecks and thermal limits.

You own the classification engine from the raw data down to the bare metal. We do not want someone who blindly throws deep learning at everything. You must look at the raw FFT data and think through the physics first.

If an object hovers with zero bulk velocity but shows a high-frequency micro-Doppler signature, it is a drone. If it has a rhythmic, low-frequency flutter, it is a bird. You build fast math heuristics to filter the obvious targets. You save the heavy ML inference for the hard cases.

Once your model works, you crush it. You own the INT8 quantization, the pruning, and the memory optimization. No Python scripts thrown over the wall - you write the C++ to deploy your logic directly onto the MCU or neural accelerator.

Degrees do not matter. Show us what you have physically deployed on constrained edge devices. Explain how you break down a classification problem before you write a single line of PyTorch. If you only train massive models on AWS with clean datasets, do not apply. Show us how you stripped a heavy model down to fit on a cheap chip without destroying accuracy. You will take fresh micro-Doppler logs from a field test on Tuesday, refine your math, retrain the model, and flash the seeker on Wednesday.

You must work in our Kyiv lab every day. We go from the bench to a live field test in days. We provide a top-of-market base salary, founding equity, a relocation bonus, company housing, and a daily allowance. We handle the logistics and buy whatever gear you need instantly. You focus 100% on the code and the hardware.