Summary written by AI from the employer’s posting. The full original text is further down.
About the role
Formulate sequential decision-making problems on process models.
Build simulation environments for reinforcement learning.
Develop and evaluate reinforcement learning methods.
Compare learned policies with appropriate baselines.
Own experiments end to end, from research hypotheses to training and evaluation.
Work across data, systems, and product to turn research into deployed capability.
Requirements
- Strong foundations in reinforcement learning, particularly in simulation and agentic environments.
- Hands-on experience implementing, training, and evaluating reinforcement learning algorithms.
- Strong research engineering and communication skills.
- Proficiency in Python and modern ML tooling.
- Ability to explain research ideas, results, and limitations clearly.
- Comfort working independently on open-ended and underspecified problems.
- Ability to move fluidly between research ideas and practical implementation.
Benefits and perks
- £90,000-£240,000 base/bonus + meaningful equity.
- Leading equity-weighted total compensation.
- Generous learning and tech budgets.
- Quarterly in-person whole-team retreats.
- Recognition and reward for output.
Skills
Full job description
Description
About HASH
We’re building toward world models from a radically different perspective to most industry labs. If you want to work on ambitious, original research with the chance to influence both scientific direction and the company itself, we’d love to hear from you.
Our founders include Joel Spolsky (former founder of Stack Overflow; acq. $1.8bn, and Trello; acq. $425m), Jude Allred (former CTO of Glitch) and Dei Vilkinsons (former CEO of Soho Strategy).
About the role
What you'll work on
- Formulate sequential decision-making problems on process models, defining observations, actions, rewards, constraints and evaluation criteria
- Build simulation environments for reinforcement learning, connecting policy decisions to process execution and computing rewards from simulated outcomes
- Develop and evaluate reinforcement learning methods that learn control policies through interaction with these environments
- Compare learned policies with appropriate baselines, including heuristics, optimisation methods and model predictive control
- Own experiments end to end, from research hypotheses to training and evaluation, measuring improvements in process performance and computational cost
- Work across data, systems, and product to turn research into deployed capability
- Contribute to research direction, technical strategy, and long-term model roadmap
- Help define the culture and standards of a high-talent, research-driven founding AI Lab team
Why HASH
- Work on first-principles AI research tied to a large and important real-world mission
- Join at a moment of rapid growth, with outsized scope and influence
- Help build a leading frontier AI lab in Europe with global ambition
- Work alongside founders who have previously built and exited businesses worth billions of dollars
- Be part of a team that moves fast, is enthusiastically open source, and is in it for the long-term
Requirements
- Strong foundations in reinforcement learning, particularly in simulation and agentic environments, as well as sequential decision-making, demonstrated through research publications, open-source contributions or applied work
- Hands-on experience implementing, training and evaluating reinforcement learning algorithms, with careful experimental design and useful baselines
- Strong research engineering and communication skills: proficiency in Python and modern ML tooling, and the ability to explain research ideas, results and limitations clearly
- Comfort working independently on open-ended and underspecified problems, translating operational needs into research objectives and adapting as priorities evolve
- Ability to move fluidly between research ideas and practical implementation: comfort quickly moving from concept to early-proof right through to release and productization
Nice-to-have
- Published research papers at leading venues, such as IROS or ICRA
- Hands-on experience with post-training techniques, particularly reinforcement learning for fine-tuning large language models
- Hands-on experience with world modelling, including using world models as simulation environments to train reinforcement learning agents
- Hands-on experience with any of the following: discrete-event simulation, process modelling, Petri nets, optimal control, operations research, combinatorial optimisation
Benefits
We meet together quarterly in-person for whole-team retreats, provide engineers with generous learning and tech budgets, and recognize and reward output.
This role pays £90,000-£240,000 base/bonus + meaningful equity.
Opens HASH’s listing on Adzuna in a new tab.