Research Engineer - LLM Training & Alignment Systems

About the role

Research, prototype, and build core infrastructure, tooling, and platforms to support the full lifecycle of large foundation model development, including data curation, model training, alignment, and evaluation, with a strong focus on scalability, efficiency, and research impact.

Key responsibilities

• Design and implement systems and workflows for supervised fine‑tuning (SFT) data curation, deduplication, and synthetic data generation to provide high‑quality training signals for large language models.
• Develop and optimise distributed training and alignment pipelines—including supervised fine‑tuning, reward modelling, and reinforcement‑learning‑based preference optimisation (e.g., PPO, GRPO)—across heterogeneous hardware platforms.
• Build and evaluate LLM benchmarking frameworks to assess model quality, alignment, robustness, and regression across training iterations.
• Collaborate with systems, hardware, and research teams to integrate novel algorithms and software frameworks into in‑house platforms, addressing performance modelling, resource allocation, scheduling, fault tolerance, and communication efficiency.
• Work with leading industry and academic experts worldwide, contribute to research publications, and drive prototype systems and patentable inventions that advance large‑scale model training and serving.

Required profile

• Hands‑on experience with large language model training and alignment, including supervised fine‑tuning, reward modelling, and reinforcement‑learning optimisation, with a solid understanding of stability, scalability, and efficiency trade‑offs.
• Strong background in large‑scale distributed training systems, optimising performance, resource utilisation, and reliability across multi‑node, multi‑device, and heterogeneous hardware environments (GPU, NPU).
• Experience building SFT data pipelines, including semantic deduplication and synthetic data generation, and understanding how data quality impacts model behaviour and alignment.
• Experience designing or applying LLM evaluation and benchmarking frameworks, including automated evaluation, preference‑based assessment, and regression analysis.
• Proficiency in Python, C/C++, or Go, with the ability to translate research ideas into scalable, reproducible prototype systems.

Required skills

• Python
• C/C++
• Go