Full-Time
Reinforcement Learning Engineer
Updated on 8/1/2026
Hammerhead AI
11-50 employees
Orchestrates AI workloads within power budgets
No salary listed
Redwood City, CA, USA
In Person
 AI & Machine Learning (1) |
|---|
Get referred to Hammerhead AI
See people who can refer or advise you
- Proven experience developing and implementing reinforcement learning algorithms, demonstrated through publications in top conferences such as NeurIPS, ICML, or ICLR, open-source contributions, or shipped products.
- At least 3 years of experience applying reinforcement learning to real-world problems, preferably in industrial automation, robotics, autonomous vehicles, energy systems, or other physical systems.
- Deep proficiency in Python and modern machine learning frameworks such as PyTorch, Jax, or TensorFlow.
- A Master of Science or PhD in Computer Science, Robotics, Operations Research, or a related field with a focus on machine learning or control theory.
- A strong theoretical background and the ability to apply reinforcement learning theory to the constraints of physical, real-world systems.
- Design and implement advanced reinforcement learning algorithms, including multi-agent reinforcement learning, model-based reinforcement learning, and deep reinforcement learning, for real-time control of data center infrastructure.
- Build and train reinforcement learning agents that can generalize to real-world physical systems.
- Lead the transition of reinforcement learning models from research and simulation to live deployment within the ORCA platform, ensuring stability and performance on mission-critical hardware.
- Analyze agent performance to continuously improve control strategies for peak shaving, workload shifting, and thermal management.
- Partner with platform engineers to define the application programming interfaces, data telemetry, and infrastructure needed to support and scale reinforcement learning agents across a global portfolio of data centers.
- Experience applying reinforcement learning in industrial automation, robotics, autonomous vehicles, energy systems, or other physical systems.
- Experience from a leading industrial or academic reinforcement learning laboratory.
- Experience with simulation platforms and reinforcement learning libraries such as Ray RLlib or Isaac Gym.
Hammerhead AI builds the ORCA platform, a software system that uses reinforcement learning to dynamically orchestrate AI workloads inside data centers within the available power budget. It optimizes the entire data-center stack—power generation, cooling, IT infrastructure, and AI jobs—to improve system-wide token productivity rather than just power efficiency. By reallocating stranded power from underutilized capacity (data centers often run at 30–50%), ORCA can boost AI output by up to 30%, helping operators deploy workloads faster and improve margins. The company targets AI factory operators, colocation providers, hyperscalers, and enterprises, differentiating by treating power as the bottleneck and delivering end-to-end power-aware orchestration to monetize idle capacity.
Company Size
11-50
Company Stage
Seed
Total Funding
$10M
Headquarters
Redwood City, California
Founded
2025
Get referred to Hammerhead AI
See people who can refer or advise you
Simplify's Take
What believers are saying
- ORCA converts stranded power into AI-ready capacity, boosting output by 30% without waiting for grid expansion.
- Each unlocked megawatt generates $20–50 million annual revenue for operators in constrained markets.
- Full system optimization of GPUs, chillers, and cooling delivers holistic token productivity beyond GPU-only solutions.
What critics are saying
- Schneider Electric or Digital Realty may deploy ORCA internally, then replicate it as a free feature, erasing revenue.
- Nvidia’s Inception Program may push competing rack orchestration tools to hyperscalers, bypassing Hammerhead as vendor.
- Google and Microsoft will roll out proprietary RL power orchestration by 2027, making ORCA obsolete for their factories.
What makes Hammerhead AI unique
- ORCA platform converts stranded power into AI capacity, boosting token output by 30% without grid expansion.
- Optimizes GPUs, chillers, cooling, and generation for holistic system-wide token productivity, not just efficiency.
- Founders orchestrated 8,000+ MW across 20 countries at AutoGrid, leveraging proven grid optimization for AI.
Help us improve and share your feedback! Did you find this helpful?
Benefits
Health Insurance
Dental Insurance
Vision Insurance
Stock Options
Company Equity
401(k) Retirement Plan