Parallax
Careers

Help machines see the world as it happens.

We are a research lab and an engineering team building perception that runs inside the frame, on the device. We hire people who like hard constraints.

Why join

The work is hard in a useful way.

Real time is a narrow target. Hitting it takes models, compilers and sensors designed together.

A problem with a clock on it.

Every model we ship has to answer inside a frame. That constraint makes the research harder and the results easier to trust.

Work that leaves the lab.

What you build runs on cameras in warehouses, on robots and in glasses. You will see it on real devices, not only in a paper.

A small team with a wide surface.

Researchers, compiler engineers and hardware people sit together. Most changes touch two of those, so you learn all three.

How we work

Calm, written, measured.

A few habits keep a small team fast without making it loud.

Measure on the device.

A result counts when it holds on the target hardware. Every change runs against the frame budget before it merges.

Write it down.

Decisions start as short written notes. Meetings are for the parts a note cannot settle.

Own the whole path.

The person who trains a model also sees it through the runtime and onto a device. Handoffs are where latency hides.

Protect deep work.

Mornings are quiet by default. Everyone gets dedicated research time each quarter to follow an idea that is not on the plan.

The practical part

Good tools, real hardware and time to think.

Competitive salary and equity, health cover in both labs, and a hardware budget so you can test on what you ship to.

2 labsZurich and San Francisco, with remote roles where the team allows
8 msThe budget every change is measured against
Your benchThe devices you ship to on your desk, not only a simulator
Time to thinkResearch time each quarter, outside the roadmap
Open roles

7 open roles across four teams.

Open a role to read what you would do and what helps. Nothing fits? Send a general application.

Research

Models that fit a frame budget
Research Scientist, DepthZurich or San Francisco

Train monocular and stereo depth models that hold their accuracy at a few milliseconds per frame on a phone-class NPU.

What you will do

  • Design depth architectures with the latency budget as a first-class constraint
  • Build evaluation sets from real deployments, not just public benchmarks
  • Publish when the result is useful to the field, and ship it either way

What helps

  • Published or shipped work in depth, 3D or dense prediction
  • Comfort with PyTorch and with reading a profiler trace
  • A habit of asking what the model costs per frame
Apply for this role →
Research Engineer, TrackingZurich

Turn per-frame detections into stable identities across occlusion, motion blur and camera shake.

What you will do

  • Own the tracker that sits between Depth models and the policy layer
  • Build tooling that replays field recordings frame by frame
  • Work with Runtime engineers to keep state on the device

What helps

  • Experience with multi-object tracking or visual odometry
  • Strong Python and working C++
  • Patience for edge cases that only appear at 3 a.m. on a loading dock
Apply for this role →

Runtime

The engine under every frame
Compiler Engineer, NPU backendsZurich or remote in Europe

Lower our graphs onto new accelerators and find the last millisecond in each one.

What you will do

  • Add and tune backends for mobile, automotive and embedded NPUs
  • Write the scheduling passes that split one frame across many models
  • Keep the benchmark suite honest across every target we support

What helps

  • Experience with MLIR, TVM, XLA or a vendor toolchain
  • Solid C++ and an interest in memory layouts
  • Some time spent reading accelerator documentation that was wrong
Apply for this role →
Systems Engineer, RuntimeSan Francisco

Make the runtime start fast, use little memory and never drop a frame under load.

What you will do

  • Own camera ingest, buffer management and the zero-copy path to the NPU
  • Build the watchdogs that keep a deployed device healthy for months
  • Profile on real hardware every week

What helps

  • Systems programming in C++ or Rust on Linux or Android
  • Experience with camera stacks, V4L2, or media pipelines
  • Care for predictable latency over peak throughput
Apply for this role →

Product

Studio, SDK and the people using them
Product Engineer, StudioZurich or San Francisco

Build the desktop app where teams connect cameras, swap models and read traces frame by frame.

What you will do

  • Ship features in Parallax Studio from first sketch to release
  • Design trace views that make a 40 ms spike obvious
  • Talk to the engineers who use Studio every day

What helps

  • Strong TypeScript and experience with a desktop or real-time UI
  • Taste for dense, calm interfaces
  • Interest in video, graphics or developer tools
Apply for this role →
Developer Relations EngineerSan Francisco or remote in North America

Help teams get from install to their first useful detection, and bring what they learn back to us.

What you will do

  • Write guides, samples and reference apps for the SDK
  • Pair with partner teams during their first deployment
  • Turn repeated questions into fixes in the docs or the product

What helps

  • Engineering experience with computer vision or embedded devices
  • Clear technical writing
  • Enjoyment of a bench full of cameras and cables
Apply for this role →

Hardware

Sensors and reference devices
Hardware Engineer, SensorsZurich

Choose and characterise the sensors and optics our models are co-designed with.

What you will do

  • Run sensor evaluations for noise, rolling shutter and low light
  • Build the reference rigs we use to collect training data
  • Work with research on the ISP settings each model expects

What helps

  • Experience with image sensors, optics or camera bring-up
  • Comfort in a lab and with a scope
  • Interest in how sensor choices change model accuracy
Apply for this role →
Apply

Tell us what you want to build.

No cover letter needed. A few sentences and a link to your work tell us more than a CV.

We use what you send only to consider your application. See the privacy policy.