Algorithms Engineer, Power Applications

Karman•Ann Arbor, MI

1d•$130,000 - $180,000•Remote

About The Position

Karman is a fast-growing NVIDIA-backed AI company enabling AI data centers to dynamically orchestrate power and unlock more compute capacity from existing energy infrastructure. For over a decade, we have applied AI to the electric grid, bringing real-time visibility and power-flow control to complex energy infrastructure. Our Karman platform, built on a custom NVIDIA module, brings that same capability to AI data centers, giving operators a way to better use the power already available to them. The Power Applications team takes power orchestration algorithms from idea to production. We develop, ship, and maintain the applications that run on Karman to orchestrate closed-loop control actions and deliver quantifiable customer value. This is a hands-on, end-to-end role. As a Senior Algorithms Engineer, you will develop and deliver these control algorithms, working day to day with the principal engineer who owns our control architecture. You will develop the theory behind our algorithms, validate them against real hardware and realistic workloads, implement them in production code alongside our software engineers, and stay with them in the field, measuring, tuning, and improving them release after release. Your time will be split between improving the algorithms already in production and R&D for the next iterations of the product, from smarter forecasting and online learning to orchestration across larger portions of a data center and new generations of GPU hardware. Strong candidates come from many paths. Some lean toward control theory and estimation, others toward building and hardening real-time systems. Both are welcome. What we require is rigor, a systems mindset, and the drive to see your work shipped and owned.

Requirements

Bachelor's degree in electrical engineering, computer science, applied mathematics, physics, mechanical or aerospace engineering, or a related quantitative field, or equivalent practical experience
Deep expertise in control systems and mathematical optimization, including feedback control, state estimation, and the practical realities of latency, noise, and missing data, ideally in distributed or edge settings
A track record of taking algorithms from concept to production in physical systems, owning their performance once deployed, and designing them to fit the whole system rather than optimizing components in isolation
Strong Python skills for prototyping and analysis (e.g. pyspark, tensorflow, pytorch) and the ability to work in production codebases in a systems language (we use Rust, and will support you in learning it)
Comfort in a continuous delivery environment: working directly in production repositories, writing tests, reviewing code, and using field performance data to understand and improve algorithms
Excellent written communication and the habits of a strong remote teammate: proactive updates, thorough documentation, empathy, and assuming good intent

Nice To Haves

Advanced degree (MS or PhD) in control, electrical or mechanical engineering, applied mathematics, robotics, computer science, or a related field, or similar research experience
Background in power systems or data center electrical infrastructure, including protection, power quality
Experience with time series forecasting, or learning-based control, including online learning and re-enforcement learning frameworks
Experience with GPU power management (NVML, DCGM, or BMC/Redfish) or with LLM inference serving and its telemetry
Experience with LLM inference stacks (e.g vLLM, Dynamo)
Publications or open-source contributions in state estimation, optimization, or control
Experience driving model development workflows in DataBricks, including exploratory data analysis, feature extraction, and model prototyping

Responsibilities

Design, analyze, and deliver improvements to the control and orchestration algorithms that allocate and enforce power across servers, racks, and rows at high speed and with hard reliability guarantees
Own algorithms end to end: theory and prototyping (Python), production implementation (Rust and Python), validation, deployment, and ongoing tuning in the field
Help to define how we measure algorithm performance, from adherence to electrical limits to impact on compute throughput, and build the benchmark and regression suites that hold each release to that bar
Continuously improve the performance, robustness, and operability of the algorithm-driven services already in production
Drive R&D that de-risks and shapes future product iterations (forecasting, automated parameter tuning, orchestration at larger scales, new compute and actuation hardware), running experiments in our lab and pilot environments and turning results into clear decisions