Research Engineer, Pretraining Scaling

Humanloop; Greater London; On-site; GBP 57,000 - 73,000; Full time

Benefits offered by this job

Visa sponsorship where possible

Generous vacation & parental leave

Equity donation matching

Job summary

Anthropic is seeking a research engineer to join the ML Performance and Scaling team in London, working on production pretrained models and performance optimization.

You will own parts of the training pipeline, debug across the stack, and design experiments to improve efficiency and uptime while collaborating with teams in London and San Francisco. The role requires in-office presence in London several days per week and requires strong communication under pressure.

Qualifications

• Hands-on experience training LLMs or deep expertise with JAX/TPU/PyTorch.
• Ability to work across research and engineering with a roughly 50/50 split.
• Bachelor's degree or equivalent required.
• Interest in on-call production systems and high-pressure launches.

Responsibilities

• Own production pretraining pipeline components, including operations and observability.
• Debug complex issues across the full stack—from hardware to training dynamics.
• Design experiments to improve training efficiency and uptime.
• Respond to on-call incidents during model launches and coordinate solutions.
• Build production logging, monitoring dashboards, and evaluation infra.
• Collaborate with teams in SF and London and with specialist groups.
• Document systems, debugging approaches, and lessons learned.

Skills

LLM training experience

JAX

TPU

PyTorch

distributed systems

debugging

on-call readiness

cross-time-zone collaboration

Education

Bachelor's degree or equivalent

Tools

PyTorch

Job description

Salary: £57,000 - 73,000 per year

Requirements:

• We want someone with hands-on experience training large language models, or deep expertise with JAX, TPU, PyTorch, or large-scale distributed systems.
• We prefer people who genuinely enjoy both research and engineering work, with an ideal split of roughly 50/50.
• We need someone who is excited about being on-call for production systems, working long days during launches, and solving hard problems under pressure.
• You should thrive when working on whatever is most impactful, even if priorities change day to day based on the production models needs.
• We value strong debugging skills across multiple layers of the stack, especially when problems are complex and ambiguous.
• You should communicate clearly and collaborate effectively, particularly across time zones or during high-stress incidents.
• We want someone passionate about the work itself and eager to refine their craft as a research engineer.
• We care about candidates who understand the societal impacts of AI and responsible scaling.
• A bachelors degree or equivalent combination of education, training, and/or experience is required.
• Your background should be in a field relevant to the role, demonstrated through coursework, training, or professional experience.

Responsibilities:

• We own critical aspects of our production pretraining pipeline, including model operations, performance optimization, observability, and reliability.
• We debug and resolve complex issues across the full stack, from hardware errors and networking to training dynamics and evaluation infrastructure.
• We design and run experiments to improve training efficiency, reduce step time, increase uptime, and enhance model performance.
• We respond to on-call incidents during model launches, diagnose problems quickly, and coordinate solutions across teams.
• We build and maintain production logging, monitoring dashboards, and evaluation infrastructure.
• We add new capabilities to the training codebase, such as long context support or novel architectures.
• We collaborate closely with teammates across San Francisco and London, as well as with Tokens, Architectures, and Systems teams.
• We contribute to the teams institutional knowledge by documenting systems, debugging approaches, and lessons learned.

Technologies:

• AI
• Hardware
• Support
• PyTorch
• LLM
• Model Training
• Quant

More:

We are Anthropic, a public benefit corporation headquartered in San Francisco, and our mission is to create reliable, interpretable, and steerable AI systems that are safe and beneficial for our users and for society as a whole. Our ML Performance and Scaling team works on training our production pretrained models and operates at the boundary between research and engineering, with deep involvement in performance optimization, hardware debugging, experimental design, and launch coordination. This role requires working in‑office 5 days per week in London, and we expect all staff to be in one of our offices at least 25% of the time. We offer competitive compensation and benefits, optional equity donation matching, generous vacation and parental leave, flexible working hours, a lovely collaborative office space, and visa sponsorship support where possible. We are a highly collaborative team that values communication, impact, and high-quality work, and this role offers extraordinary learning opportunities working alongside world‑class researchers and engineers on some of the largest training runs in the industry.

last updated 36 week of 2026