3 hours ago
London, United KingdomSenior
Responsibilities
- Own the product capability and roadmap for improving LLMs based on customer workflows.
- Build reproducible post-training harnesses covering data preparation, training, evaluation, model artifacts, and serving hand-off.
- Apply supervised fine-tuning, preference optimization, RLHF/RLAIF, reinforcement learning with verifiable rewards, distillation, rejection sampling, and related methods.
- Develop synthetic-data and expert-data pipelines with quality, provenance, licensing, and contamination controls.
- Design reward signals, verifiers, and rollout workflows while detecting reward hacking, benchmark overfitting, model collapse, and regressions.
- Improve open models against agent and engineering benchmarks and determine whether training or product changes provide better results.
- Make post-training available through managed workflows and expert-facing interfaces.
- Keep customer data and model artifacts within customer-controlled environments across SaaS, VPC, and air-gapped deployments.
- Collaborate with Research and platform teams on diagnostics, benchmarks, production serving, and model routing.
- Help shape the future team and capability as the product develops.
Requirements
- Demonstrated ability to measurably improve model behavior through post-training and deploy the resulting capability in production.
- Hands-on experience across data, training, evaluation, debugging, and deployment of model artifacts.
- Strong understanding of modern post-training methods and how to select them based on task and evidence.
- Experience creating high-quality training datasets from expert, behavioral, or synthetic sources.
- Ability to diagnose instability, drift, reward exploitation, leakage, and evaluation overfitting.
- Product-minded approach to defining user workflows, success criteria, trade-offs, and operable experiences.
- Strong Python and PyTorch or equivalent model-training framework skills.
- Experience with distributed training, accelerator utilization, checkpointing, recovery, experiment tracking, and reproducibility.
- Understanding of model and agent evaluation, controlled experiments, and ablations.
- Knowledge of data governance, privacy, provenance, and safe handling of customer-owned data and artifacts.
- Experience with model packaging and the operational path from training through evaluation to production serving.
- Clear technical writing and communication across research, product, engineering, delivery, and customer teams.
- Preferred experience includes post-training open-weight models for tool use, reasoning, or long-horizon agent workflows.
- Preferred experience includes large-scale rollout generation, reward models, verifiers, online reinforcement learning, managed model customization products, open-source training frameworks, expert workflows, or regulated and air-gapped environments.
Benefits
- The role supports SaaS, VPC, and air-gapped deployment environments.
- PhysicsX is headquartered in the United Kingdom, with offices in London, New York, Singapore, and an expanding Bay Area presence.
- The company is committed to equal employment opportunity and supports diversity and inclusion initiatives.
Categories
About PhysicsX
PhysicsX builds an AI-driven simulation software stack that enables high-fidelity, multi-physics modeling and optimization for engineering and manufacturing teams. The company sells software and delivery services to enterprises in aerospace and defense, automotive, semiconductors, materials, and energy. It is privately held and headquartered in London, with offices in London and New York.