1 day ago
Helsinki, FinlandSenior
Responsibilities
- Apply quantisation, pruning, and knowledge distillation to real models and recover accuracy through fine-tuning, retraining, or other methods.
- Profile model memory, bandwidth, and compute usage and convert findings into optimization decisions.
- Optimize models for hardware accelerators including GPUs, NPUs, and TPUs.
- Deploy production-ready models in client environments and QuTwo OS.
- Build domain adaptation harnesses involving retrieval pipelines, tool integrations, structured prompting, or agent loops.
- Work with researchers, engineers, and client teams to move research systems into production.
- Use Python for research code and model artifacts and C/C++ for low-level optimization.
Requirements
- Practical experience applying model compression techniques such as quantisation, pruning, and knowledge distillation.
- Experience profiling models and optimizing them for specific hardware accelerators.
- Fluency with current open-weight model families and their tooling.
- Experience building domain adaptation harnesses and deciding when they are preferable to fine-tuning.
- Proficiency in Python and C/C++.
- Several years of hands-on experience shipping models to production in industry or an applied research lab.
- Ability to work across changing projects and hardware stacks and own work through deployment.
Categories
About Qutwo
Qutwo builds enterprise AI and quantum-computing solutions, combining its Qutwo OS, ML pipelines for neural network compression, and security-focused SaaS services for industrial optimization, simulation, and analytics. The company partners with enterprises to develop classical, hybrid, and quantum systems and offers software plus applied research engagements. It is privately held, Europe-based, and was co-founded by pioneers behind IQM and Silo AI, with backing from founders of Skype, Hugging Face, Quantinuum, and Supercell.