Machine Learning Engineer- Reinforcement Learning

London (hybrid, 1 day/week in Kings Cross) - Solve Data Centres Cooling issues. Cooling is one of the largest items on a data centre's energy bill, and most sites run it conservatively because getting it wrong puts the hardware at risk. Our client trains reinforcement learning agents to control cooling systems on live sites, cutting cooling energy without breaching the temperature and humidity limits operators are contractually bound to. They're hiring an ML Engineer - Reinforcement Learning to build those agents and get them running on real data centres. You'll report to the CTO / Head of AI and work across the line between research and deployment.

The System: The agents don't learn on the live plant. They train against a digital twin of each site, then move to production once they're safe. Reward and constraint design is shaped by ASHRAE standards and customer SLAs - air temperature, humidity, and rate-of-change limits on cooling air and chilled water setpoints. Training is federated across multiple sites. Agents share learned control strategies without any site's operational data leaving the building, which delivers significantly more savings than a single-site approach. Models are deployed on-prem at the edge, then monitored and retrained in place.

What You'll Own:

• Reinforcement Learning: Train and deploy deep RL agents for live cooling control. Design reward functions and constraints that hold up against physical limits and SLAs, not just in a notebook. Move between research-style exploration and the engineering work to make something stable on a real site.
• Simulation and Digital Twins: Build and improve the physics-based simulators, surrogate models, and digital twins the agents train against. Close the gap between what works in simulation and what holds on real hardware.
• Production and Deployment: Federated and distributed training across sites. Edge deployment, monitoring, and retraining of agents already running in production.

What We're Looking For:

• Essential: 3-5 years training and deploying deep RL agents in Python, PyTorch or JAX, and RL libraries such as Gymnasium. A background in physical systems - engineering (mechanical, electrical, structural, biomedical), physics, robotics, autonomous driving, or control systems - and the instinct to reason about what's physically possible, not only what's mathematically possible. Comfortable iterating between research exploration and the engineering needed to run on a live site. A degree in engineering, CS, or physics.
• Useful: Control systems (classical control, MPC), HVAC, thermodynamics, power systems, or data centre operations. Federated learning, distributed training, or edge ML deployment. Simulation experience - building or using physics-based simulators, digital twins, surrogate models, or large physics models. Published research or open-source contributions.

Who You Are: You want both halves of this job. You'll run experiments and read papers, but you also want your work controlling real equipment, with the constraints that come with that. RL experience limited to advertising or multi-armed bandits won't carry over here - the physical world doesn't behave like a recommendation system. A pure maths or CS background with no feel for physical systems will struggle, and so will anyone after a pure research seat or a pure production one. This sits in the middle.

What's On Offer: £110K-£150K, plus competitive equity. A genuine technical problem: RL on physical systems, under real constraints, deployed on live infrastructure. Direct access to the CTO and founding team. Hybrid working, one day a week in the Kings Cross office. Visa sponsorship available on a case-by-case basis. Get in touch for a confidential conversation.