Edge AI Engineer (m/w/d)

Your mission:
Build AI that runs in the real world. On real robots. Under real constraints. 

At Autonomous Teaming, we build autonomous robotic systems operating in extreme, GPS-denied environments. Our models run fully on edge hardware (Jetson, FPGA, custom boards), with no cloud, no fallback, no excuses. 
We’re looking for an engineer who loves hard problems : real-time inference, low-latency pipelines, CUDA kernels, TensorRT graphs, and deploying ML models directly on hardware. 
If you enjoy debugging things that only break on the robot, this role is for you. 
 
Missions : 

Own the full pipeline from model to real-time inference on embedded devices: 
• Optimize deep neural networks for Jetson, FPGA or ARM boards 

• Apply quantization, pruning, distillation to hit strict FPS, power and memory budgets

• Convert & compile models using TensorRT, ONNX, CUDA, C++

• Build ROS nodes integrating optimized perception into the full robotic system 

• Debug runtime failures, memory leaks, thermal throttling, kernel-level issues

• Benchmark and validate performance directly on hardware 

• Ship models that run reliably in real-world, harsh environments

Your profile:
Must-have

• Strong experience in CUDA & C++
• Hands-on work with TensorRT, ONNX, TVM or similar compilers
• Practical experience with quantization/ pruning/ INT8 / FP16
• Experience deploying models on Jetson/ embedded GPUs/ ARM / FPGA
• Comfortable with PyTorch, Python, Linux, Git
• Engineer mindset : measurement→ optimization→ validation

Nice-to-have 

• ROS (building nodes, integrating perception stacks)
• Custom accelerators, DSPs or hardware-specific toolchains
• Profilers : Nsight, perf, tegrastats, TensorRT profiler
• Experience in robotics , autonomous systems, aerospace, automotive or defense

Why us?:
• You ship code that directly controls real robots
• You work on constrained hardware, where every millisecond and every watt matters
• You solve problems that cloud ML engineers never face
• You own your optimizations end-to-end : from model to field deployment
• You work in a small, high-performing team where ownership is real

If you want a job with clean layers and abstract diagrams, this is not it.