ML Engineer, Model Compression

Join QuTwo, a European AI lab for the quantum era, on our quest to bridge classical AI and quantum computing systems.

A trained model is only half the story. The other half is making it run fast, lean and reliably on the hardware where it actually has to live. As an ML Engineer you will take models from our research team and clients and turn them into efficient, production-ready systems.

You will join a small team that cares about every millisecond and megabyte. If you are a humble learner, curious about what sits beneath the framework, we encourage you to apply.

Role Description

You will sit where research becomes something people can use. You take trained models from our researchers and clients and make them run within real memory, latency and hardware limits, all the way to deployment in client environments and in QuTwo OS, the intelligence layer for industrial AI.

You will work closely with our researchers, engineers, and client teams in a fast-moving environment, taking initiative to bring research into production.

• Model compression in practice. You have applied quantisation, pruning and knowledge distillation to real models, and know how to recover accuracy afterwards through fine-tuning, retraining or other healing methods.
• Profiling, not just benchmarking. You can find where a model spends its memory, bandwidth and compute, and turn that into concrete optimisation decisions.
• Hardware-aware optimisation. You have optimised models for specific accelerators such as GPUs, NPUs or TPUs, and understand the trade-offs that come with each.
• Open-weight models. You are fluent with today's open-weight model families and the tooling around them.
• Domain adaptation harnesses. You have built the scaffolding that makes a general model useful in a specific domain, such as retrieval pipelines, tool integrations, structured prompting or agent loops, and know when that beats fine-tuning.
• Python and C/C++. Python to work with research code and model artefacts; C/C++ for low-level optimisation where Python overhead is not an option.
• Industry experience. Several years of hands-on work shipping models to production, in industry or in an applied research lab
• Comfort with change. You enjoy switching between projects and hardware stacks, and you take ownership from first idea to final deployment.
• Custom kernel development with CUDA, Triton, HIP/ROCm or Metal
• On-device and edge ML, for example on NVIDIA Jetson, mobile SoCs or embedded hardware
• Optimising concurrency, parallel execution and memory bandwidth for efficient inference
• Model serving and deployment at scale
• Generating synthetic data to fine-tune or heal compressed models