Language Model Scientist

The role

Our client is building a group-level AI Center of Excellence to set the technical direction for artificial intelligence across its operating companies. This role owns language modelling — but the emphasis is not frontier scale. It is domain-specific and open-weight models, made small enough and reliable enough to run on the client's own terms: on device, on premise, and within a sovereign deployment where required. Reasoning quality, and the reinforcement learning that produces it, sits at the centre of the work.

Responsibilities

• Model adaptation. Adapt open-weight models to domain tasks through fine-tuning, LoRA and PEFT, instruction tuning and retrieval augmentation.
• Small models. Produce compact models through distillation, pruning and quantisation for on-device and on-premise inference.
• Reinforcement learning. Apply preference optimisation, RLHF and RLAIF, and verifiable-reward methods to improve reasoning and reliability.
• Continuous learning. Build synthetic data generation and self-improving training loops that keep models current without full retraining.
• Evaluation. Design evaluation harnesses for reasoning, factuality, safety and regression — measurement the business can trust.
• Hardware optimisation. Tune inference cost, latency and throughput against the target silicon.

Experience

• Sustained specialisation in language modelling, with hands-on training and post-training work
• A record built at a leading technology company, a frontier AI research organisation, or a research-intensive enterprise
• Models shipped into production, with accountability for their behaviour and cost in operation
• Distributed training and modern inference stacks; strong Python and PyTorch
• The standing to guide AI teams already established within the group's operating companies

Technical areas

• Fine-tuning, LoRA, PEFT and instruction tuning
• Retrieval-augmented generation
• Reinforcement learning and RLHF
• Reasoning and inference-time compute
• Knowledge distillation and model compression

Education

A postgraduate degree in computer science, machine learning or a related discipline, from a leading institution in the United States or India. A doctorate is preferred.