AI Engineer, Foundational AI Technologies

The role

SkyDynamics develops its own AI technologies for aerospace and defence, among them a retrieval engine written in Rust and the Assurance Small Language Model, a compact model designed so that each statement it produces can be traced to the evidence it was given.

This position works below the level of the usual frameworks. You derive the method, write the numerical code in C++, and make it run in real time on the hardware at hand, with behaviour that can be measured and explained.

What you will do

• Design and train compact models for retrieval, language and time-series data.
• Write inference and training kernels in C++: tensor operations, attention, quantisation and memory layout.
• Apply linear algebra and numerical methods directly: decompositions, optimisation, and control of numerical error.
• Make models run in real time at the edge, on CPUs and embedded processors, within a fixed budget of time and memory.
• Build the evaluation: test sets, metrics and traceability from each output to its evidence.
• Work with the database and safety-critical software engineers on the interfaces to LLiDA and to the retrieval engine.
• Read the research, reproduce what matters, and publish where it helps.

What you bring

• 8 or more years of professional or research experience in machine learning or numerical software.
• Low-level C++ programming: memory management, SIMD, cache behaviour, concurrency and profiling.
• Applied mathematics: linear algebra, probability, optimisation and numerical analysis.
• A sound understanding of how neural networks work inside, beyond the use of a framework: you can implement a transformer without one.
• Python for experiments, with PyTorch or a comparable framework.
• Experience in taking a model from research to software that runs under real constraints.
• Master's degree in computer science, electrical or computer engineering, mathematics, physics or a related field.
• Greek, fluent (required).
• English, fluent (required).

Advantages

• A PhD in machine learning, applied mathematics, computational science or a related field.
• Rust, for work on the retrieval engine.
• Small language models, retrieval-augmented generation, and model compression (quantisation, distillation, pruning).
• Inference on embedded processors, GPUs or FPGAs.
• Experience in a safety-critical field, and familiarity with DO-178C, MISRA C++ and the EASA guidance on AI.
• Signal processing or time-series modelling.
• Publications in peer-reviewed venues.

Eligibility

• Right to work in the European Union.
• Military obligations completed, or a lawful exemption, where they apply to you.