Workloads you would be involved in

• Design, implement, and optimise large-scale pretraining and post-training pipelines for foundation models.
• Build efficient and scalable training algorithms for large language and multimodal models.
• Translate state-of-the-art research papers into production-grade, performant, and maintainable training code.
• Optimise distributed training workloads across multi-node and multi-GPU environments.
• Work with deep learning distributed training frameworks and optimise training efficiency, scalability, and fault tolerance.
• Design and run ablation studies to identify effective architectures, hyperparameters, training strategies, and scaling behaviors.
• Improve training throughput, memory efficiency, and hardware utilisation through algorithmic and systems-level optimisation.
• Develop and integrate training pipelines with continuous evaluation, benchmarking, and validation frameworks.
• Work on efficient checkpointing, model serialisation, export, and deployment-ready artifact generation.
• Debug and resolve training instability, convergence issues, gradient pathologies, and distributed systems failures.
• Implement and optimise post-training workflows including instruction tuning, alignment, preference optimisation, and model refinement pipelines.
• Build tooling for experiment tracking, reproducibility, monitoring, and large-scale training orchestration.
• Collaborate closely with research, infrastructure, and model teams to rapidly prototype and productionise new ideas.
• Benchmark models across quality, efficiency, latency, and resource utilisation metrics.
• Contribute to the design and improvement of internal training stacks, libraries, and developer tooling.

You are a good fit if you:

• Have 3+ years of experience in machine learning engineering, deep learning systems, or related domains.
• Have strong expertise in deep learning and modern neural network training methodologies.
• Are proficient in implementing research ideas and papers into production-efficient code.
• Have deep understanding of distributed training paradigms including data, tensor, pipeline, and expert parallelism.
• Have strong programming skills in Python and performance-oriented ML development.
• Have experience debugging and optimising complex ML training pipelines with profilers like nsys
• Understand training dynamics, hyperparameter tuning, scaling laws, and model evaluation methodologies.
• Enjoy working at the intersection of research and engineering to build scalable AI systems.
• Care deeply about performance, reproducibility, reliability, and engineering quality.

You are a strong candidate if you have experience with:

• Distributed training frameworks such as Megatron-LM (must have experience), PyTorch Distributed, DeepSpeed, FSDP, or similar systems.
• Large-scale pretraining and post-training workflows for foundation models.
• Optimisation techniques including mixed precision training, activation checkpointing, gradient accumulation, and memory-efficient training.
• Continuous evaluation, benchmarking, and experiment management systems.
• Model export and deployment with frameworks such as SGLang, vLLM or TensorRT.
• CUDA, GPU kernel development, or GPU architecture and performance optimisation.
• Profiling and debugging tools for distributed and GPU workloads.
• HPC or large-scale GPU cluster environments and training infrastructure.
• Preference optimisation, alignment methods, or reinforcement learning-based post-training techniques.
• Building internal ML tooling, training infrastructure, and reusable libraries for large-scale experimentation.