Open NEURA Robotics’ careers page and almost every engineering role carries a small parenthetical: “(human).” Fullstack Engineer (human). RT & Kernel Engineer (human). Robotics Middleware Engineer (human). It reads as an in-joke from a company whose entire product line is cognitive robots built to do physical work, and it hints at what the interview cares about. NEURA hires people to build machines that will stand next to a person on a factory line, so the screen leans less on clever puzzle-solving and more on whether you can reason about real hardware that moves and can hurt someone.
The stakes are worth understanding before you walk in. In June 2026 NEURA announced a Series C of up to $1.4 billion, which it called the largest round a full-stack robotics company has raised, with capital from NVIDIA, Qualcomm, Amazon, Bosch, Schaeffler, Tether and the European Investment Bank. The company runs around 1,400 people out of Metzingen, a small town near Stuttgart, reports an order backlog above $1 billion, and says it wants to be producing millions of robots a year by 2030. All of that turns into a flood of open reqs and a hiring process that still feels like a European deep-tech scale-up rather than a rehearsed big-tech pipeline.
Your track decides the questions
There is no single NEURA interview. The company ships a cognitive robot arm (MAiRA), a lighter collaborative arm (LARA), an autonomous mobile base for heavy loads (MAV), a household assistant (MiPA), and 4NE-1, which it bills as Europe’s first production-ready humanoid and is now on its third generation. Tying them together is the Neuraverse, an app-store-style platform where robots share learned skills, plus NEURA Gyms, physical spaces where the machines train on real tasks. The AI stack underneath is called AURA.
Because the surface area is that wide, the recruiter’s first job is figuring out which world you belong in. A control or firmware engineer gets pushed on real-time C++ and what happens when a loop misses its deadline. A middleware engineer gets ROS 2 and DDS. Someone on the NEURAGym cloud team gets AWS and infrastructure-as-code. A Neuraverse front-end candidate gets React and 3D rendering in the browser. Same company, very different afternoons.
| Engineering track at NEURA | Stack you’ll be tested on | Representative question |
|---|---|---|
| Real-time control and firmware | Modern C++ (17/20), embedded Linux, QNX or an RTOS, real-time scheduling | What runs between a sensor read and an actuator command, and where does latency hide? |
| Robotics middleware and SDK | ROS 2 (rclcpp/rclpy), FastDDS/RTPS, gRPC, language bindings (PyBind11, JNI) | Why use DDS instead of raw sockets between two nodes on the same robot? |
| Perception and vision | C++/Python, point clouds, object detection, sensor fusion, QoS tuning | How do you keep a vision pipeline inside a fixed latency budget? |
| Controls, locomotion and teleoperation | Kinematics, task-space and joint-space control, MuJoCo, low-latency pipelines | A policy trained in simulation fails on the real arm. What is your first move? |
| Cloud and NEURAGym platform | TypeScript/Python, AWS, infrastructure as code (CDK/Terraform), GitLab CI, Docker | Design the service that ships a new skill to a fleet of robots. |
| Full-stack and Neuraverse UI | React/Vue, Three.js/WebGL, WebSockets/ROSBridge, REST/GraphQL | Render live 3D telemetry from a robot in the browser without dropping frames. |
The loop, and how much to trust the shape of it
NEURA does not advertise a fixed five-round format the way a US company might, and public write-ups are thin because most of its hiring happens in Germany. Take what follows as the common pattern rather than a promise. It opens with an HR or recruiter call covering your background, why robotics, whether you can relocate, and your right to work in Germany. English is the working language and most postings only ask for strong English, with German listed as a nice-to-have.
From there you talk to the hiring manager, who goes deep on your resume and maps it to a specific team. Then a technical round that depends on the role: some candidates report a live coding session in C++ or Python, others get a take-home followed by a review where you defend your choices. A panel with future teammates follows, mixing coding, domain questions and a little system design. Senior roles often close with a lead, and at a company this founder-driven, David Reger’s org can show up late in the process. The teams work in Scrum, so questions about how you handle sprints and code reviews come up alongside the technical rounds.
What actually earns an offer
If you are on the real-time side, C++ discipline on a control loop is the thing to get right. Interviewers want to hear that you keep heap allocation out of the hot path, that RAII and move semantics are second nature, and that you know why a 1 kHz controller cares about cache misses and lock contention. Speak the language of determinism: worst-case execution time, priority inversion, jitter. A frequent opener asks you to trace the path from a sensor reading to an actuator command and point out every place latency creeps in.
Middleware and SDK candidates should treat ROS 2 as a mental model rather than a line on a resume: nodes and topics, the DDS discovery layer beneath them, and the Quality of Service settings that decide whether a message is dropped or queued under load. NEURA uses FastDDS specifically, so knowing RTPS, multicast behavior and reliability profiles helps. SDK roles pile on gRPC and the real pain of shipping client libraries across Python, C++ and Java with binding tools like PyBind11.
Perception and controls candidates get asked about the sim-to-real gap, which NEURA takes seriously enough that it partnered with Dassault Systemes to work on it. Be ready to explain why a policy that looks perfect in MuJoCo comes apart on a physical arm, and how you would close that with system identification or domain randomization. Teleoperation roles dig into low-latency dual-arm control and the difference between reasoning in task space versus joint space. System design questions tend to be grounded in the product: fleet orchestration across many robots, skill distribution through the Neuraverse, or a data pipeline that moves point clouds off a robot without saturating its link.
A handful of questions come up often enough to rehearse:
- “Walk me through your control loop from sensor to actuator, and tell me where the latency is.”
- “Why DDS rather than plain UDP between two processes on the same machine?”
- “Your model works in simulation and fails on real hardware. How do you debug that?”
- “Design the API a customer uses to load a new skill onto a robot through the Neuraverse.”
- “How do you test software when a bug can physically damage a machine or injure someone?”
The parts strong candidates underestimate
Two things sink otherwise solid engineers. The first is treating this as a pure software job. NEURA is a hardware company that happens to write a lot of software, and the people who land offers talk about the robot as the thing they are serving: what fails when a joint overheats, how the system reaches a safe state, what a wrong sign in a controller does to a moving arm. The second is the relocation reality. These roles are on-site in Metzingen, Munich, Bielefeld or Zurich, and Metzingen is a quiet outlet-shopping town, not Berlin. If you need sponsorship, the EU Blue Card is the usual route for engineers, and it carries a minimum salary threshold that changes yearly, so confirm the current number for your situation before you count on it.
On compensation, reset any US anchoring. Software pay at a well-funded German scale-up is strong for the European market but lands below Bay Area figures, and equity is structured differently from a typical Silicon Valley grant. If money is the deciding factor, get the specific band for your level and location in writing early instead of extrapolating from US data.
The engineers who get through share one habit: they can talk about physical failure without flinching. Plenty of candidates can reverse a linked list. Far fewer can explain, in plain terms, what their code does when the hardware it controls is a 60-kilogram arm moving near a person. That conversation is the one NEURA is really screening for, and it pays to walk in ready to have it.
Practice the behavioral round:
