Product ML Engineer: Open-Model AI Post-Training

PhysicsX Ltd.; Greater London; On-site; GBP 110,000 - 160,000; Full time

Job summary

PhysicsX Ltd. in London is seeking a hands-on product ML engineer to own the capability that lets customers improve LLMs for their work.

You will work closely with Research and Platform teams to turn post-training methods into deployable features, with a focus on open models performing well on our evaluation benchmarks. The role covers SFT, RLHF/RLAIF, distillation, synthetic data, and verifiers, with responsibility for data governance, reproducibility and safe handling of customer artifacts.

Qualifications

• Strong Python and PyTorch skills for ML model training and evaluation.
• Hands-on across data, training, evaluation and debugging.
• Ability to design and run controlled experiments and ablations.
• Knowledge of data governance and safe handling of customer data.

Responsibilities

• Own post-training capability from base model to measurable improvements.
• Build reliable post-training harnesses for data, training, evaluation and hand-off.
• Improve open models against internal benchmarks using evidence.
• Develop methods including SFT, RLHF/RLAIF, distillation and synthetic data.
• Create synthetic/expert data pipelines with quality filtering and provenance.
• Design reward signals, verifiers and rollout workflows.
• Make post-training available through managed workflows and expert interfaces.
• Ensure data/artifacts stay within customer environments across SaaS/VPC/air-gap.

Skills

Python

PyTorch

Experimentation

Model deployment

Tools

Distributed training

Checkpointing

Experiment tracking

Job description

PhysicsX Ltd. in London is seeking a hands-on product ML engineer to own the capability that lets customers improve LLMs for their work.

You will work closely with Research and Platform teams to turn post-training methods into deployable features, with a focus on open models performing well on our evaluation benchmarks. The role covers SFT, RLHF/RLAIF, distillation, synthetic data, and verifiers, with responsibility for data governance, reproducibility and safe handling of customer artifacts.