Research Engineer — Applied Science; ML

Bridge the gap between cutting-edge AI research and practical application. You will own the design, training,and fine-tuning of advanced machine learning models to solve complex, domain-specific problems. We value engineering driven by first-principles thinking — someone who genuinely enjoys digging into the math of thelatest arXiv papers, exploring novel architectures, and rapidly translating theoretical breakthroughs into functional, high-performance prototypes.

• Model architecture & design: research, design, and prototype deep learning architectures tailored to our data domain — generative models, geometric deep learning, large-scale transformers.
• Experimental exploration: own rigorous experimental roadmaps — ablation studies, hyperparameter optimisation, and loss-function engineering.
• Algorithmic & compute optimisation: integrate novel optimisations (custom attention, tokenisation, layer normalisation) with an eye on memory efficiency and training stability.
• Data strategy & feature engineering: design representation, augmentation, and synthesis strategies for complex, constrained biological datasets.
• Collaborative hand-off: partner with MLOps so prototypes are built for stability and transition cleanly into scaled production.

• Advanced ML & deep learning: deep theoretical and practical grasp of modern frameworks (PyTorch preferred), optimisation, and neural-network primitives.
• Scientific computing & math: strong linear algebra, calculus, probability, and statistics; fluency with Num Py, Sci Py, and Pandas.
• Research translation: read an academic paper, understand its mechanics, and reproduce or adapt it in clean, modular Python.
• Software fundamentals: maintainable Python, Git, and rigorous testing with PyTest.
• Experiment tracking: Weights & Biases, MLflow, or Tensor Board to document and reproduce dense training runs.

An advanced degree (MSc/PhD) in a quantitative field; experience with large-scale generative models orspatial, time-series, or geometric data; open-source contributions or publications at NeurIPS, ICML, ICLR, or CVPR.