Senior Machine Learning Engineer - Foundational Models

You will join the senior engineering group behind Aurora Labs, working alongside people who have shipped execution-intelligence products for automotive and mission-critical systems. This is an all-senior team: no ramp-up program, no layers between you and the model. You own your experiments end to end.

What you'll do

• Design and iterate on foundation model architectures for representing compiled programs and their execution behavior
• Design, run, and analyze training and fine-tuning experiments - from hypothesis to reproducible result
• Build and optimize data pipelines that turn large volumes of real-world execution and telemetry data into training corpora
• Develop evaluation methodology: benchmarks, regression detection, and accuracy tracking across hardware targets
• Take models from experiment to production across server, robotics, and bare-metal deployment targets
• Work directly with senior systems engineers who own the binary-analysis side of the pipeline

The work sits at a rare intersection. If the research line from learned performance prediction on compiled code (Ithemal, GRANITE, PerfVec) or tokenization of ISA-level representations means something to you, you will feel at home here. If it doesn't yet but you have deep foundation-model experience and curiosity about how software meets hardware, that works too - the low-level side can be learned from the team.

Mandatory

• Senior-level, hands-on machine learning experience - you have trained and fine-tuned transformer or other foundation-model architectures on real projects, not only used them via APIs
• Strong PyTorch (or JAX) experience and fluency with the Python data stack
• Experience building data pipelines for model training: collection, cleaning, tokenization/encoding, and throughput optimization at scale
• Rigorous experimental practice - you design evals, strengthen baselines, debug regressions, and track down data or training bottlenecks in a self-directed way
• Clear written and spoken communication of experimental results

Big plus

• Ability to read assembly / ISA-level code (x86-64, ARM, or RISC-V) - our training data is compiled binaries, and understanding what the tokens mean is a real advantage
• Working knowledge of C/C++ or Rust
• Exposure to compilers, toolchains, binary analysis, or performance engineering
• Embedded systems or robotics experience

What we offer

• A small, all-senior team where your experiments ship into a real product, not a research shelf
• Foundation-model work with a concrete, measurable target: predicting real execution behavior of real software
• Full-time position in Skopje with the ITQuarks engineering group