Member of Technical Staff, Matrix

About Midcentury

Midcentury builds data and simulation infrastructure for physical AI. Matrix is their agentic simulation platform for perception, autonomy, and robotics teams, designed to bring hardware and software stacks into high-fidelity simulation and test them at scale before deployment.

With Matrix, teams can build calibrated digital twins and testing environments, configure sensors and system behavior, and run thousands of controlled simulation scenarios in parallel. 3D agents help automate the workflow by creating and modifying environments, setting up experiments, running evaluations, and analyzing failures.

The goal is to make simulation an everyday engineering layer for physical AI: faster to set up, easier to scale, and useful across everything from fixed perception systems to autonomous machines and robotics.

The Opportunity

Midcentury is hiring a core Member of Technical Staff to help build Matrix from the ground up. You will work directly with the founders, researchers, and engineers to shape the platform's architecture and ship its core capabilities, ranging from bringing real customer systems into high-fidelity environments to building the agentic harness for creating scenarios, running experiments, and iterating automatically.

This is an early, high-ownership engineering role spanning simulation, evaluation, and platform infrastructure. You should bring deep simulation experience and be comfortable working across the surrounding platform stack, including cloud execution, parallel GPU workloads, APIs, and developer tooling.

You will take ambiguous research and customer needs, turn them into concrete technical requirements, and build systems people can reliably use, often in tight loops with customers and the rest of the Matrix team.

What You'll Do

- Build the simulation platform: develop interfaces for configuring environments, defining tasks, connecting policies, and running reproducible experiments across simulation backends

- Make evaluation useful: implement task-success metrics, scenario suites, controlled variations, and regression tests that reveal failures in manipulation, physical reasoning, and long-horizon execution

- Scale execution: build orchestration for parallel simulation and policy inference, including GPU scheduling, storage, monitoring, and recovery from failed runs

- Bring real customer systems into Matrix: translate cameras, LiDAR, radar, robot embodiments, environments, and real-world data into calibrated simulation configurations

- Build high-fidelity simulation capabilities: develop environments, rendering, sensor models, dynamics, and other components needed to reproduce how physical systems behave and perceive the world

- Model real sensing systems: build and calibrate cameras and other sensors, including optics, noise, latency, failure modes, and ISP/DSP behavior

- Connect simulation to real-world data: translate recorded tasks, environments, and failure cases into simulation scenarios; measure fidelity and document limitations

- Build the agentic harness: create APIs and tools that let agents configure scenarios, launch experiments, inspect results, and iterate with measurable feedback

- Make runs inspectable: capture synchronized observations, actions, simulator state, and evaluation traces so users can replay failures and compare model versions

- Train models with real-world physics data: use real-world interaction data and simulated trajectories to train and post-train robotic policies and world models, then evaluate improvements in physical reasoning, manipulation, and long-horizon task execution

- Own what you ship: work directly with users, debug across the stack, and make practical tradeoffs around fidelity, throughput, reliability, and cost

You Should Have

- Hands-on experience with robotics or physical-system simulation using tools such as Isaac Sim/Lab, MuJoCo, Gazebo, or similar; experience with Isaac Sim and OpenUSD is especially relevant

- Experience building, calibrating, and validating simulated sensors or physical systems, including cameras, LiDAR, radar, optics, noise, timing, or ISP/DSP behavior

- Strong intuition for sim-to-real fidelity: you can take incomplete real-world specifications or data, build a simulation model from them, and reason quantitatively about where it matches reality and where it does not

- Strong software engineering fundamentals, with excellent Python skills and experience building production systems, tooling, APIs, and performance-sensitive pipelines

- A track record of shipping reliable capabilities, not just research prototypes

- Comfort operating from ambiguity: you can turn incomplete requirements into working systems, make sensible technical tradeoffs, and communicate assumptions and limitations clearly

Nice to Have

- Experience with scalable compute, including GPU workloads, cloud infrastructure, distributed execution, scheduling, or systems for running large numbers of experiments in parallel

- Isaac Sim/Lab, MuJoCo, OpenUSD, or comparable simulation and rendering systems

- Sensor simulation, calibration, real-to-sim pipelines, and perception/autonomy testing

- Experience simulating or calibrating sensing modalities beyond standard RGB cameras, including thermal/IR, radar, LiDAR, UV, or other non-visible sensors

- GPU workloads, cloud orchestration, C++/Rust, distributed systems, or simulation performance optimization

- Agent tooling, evaluation harnesses, SDKs, or interfaces for visualizing and replaying experiments

- Based in NYC or open to relocating