Staff Software Engineer, Computer Vision

About Chip

Chip is the first mass-market American robot. He just happens to be your car.

Chip is building a new category of vehicle. Most cars are engineered for highway
speeds and long-distance range. The reality is that 60% of daily trips are under
6 miles. Chip is built for those trips: fully electric, street-legal on roads up
to 35 mph, 100+ mile range, and made to fit the way people actually move through
their neighborhoods.

We're also launching Chip Go!, a remote-driving service that keeps your
vehicle moving when you're not inside it. Say the word and he'll go off to park
himself, run an errand, get cleaned, or come get you and the family. Rolling out
regionally.

We launched publicly in July 2026 and reservations are open nationally.

The role

Chip Go! is remote-driven today. When your Chip goes to park himself or run an
errand, a person somewhere else is driving him.

You own how Chip sees. What's around him, what's a person, what's a hazard, and
what's someone trying to take him. That's three problems sharing a camera and
almost nothing else: driving him without a driver, knowing where people are, and
watching him when nobody else is.

You'll report to the C-suite and work closely with the vehicle team, who own the
drive-by-wire and networking stacks.

The stack

Rust on the vehicle, running Linux on an industrial PC. Cameras inside and out,
hardware-accelerated encode, and actuation is drive-by-wire over CAN.

Inference runs at the edge on a Jetson, on the vehicle. Nothing round-trips to a
server to decide whether to stop. Your latency and power budgets are real
numbers on real silicon rather than a line on a cloud bill, and what fits on
that board is a design constraint from day one rather than something to optimize
later.

You're not starting from a blank vehicle. The hard, unglamorous parts (cameras
that survive a road trip, video that arrives, a car that takes commands) already
work. What's missing is everything that turns pixels into a decision.

What you'll do

• Own perception end to end. From the sensor on the vehicle to the signal something acts on, including the parts that only break in rain, at dusk, or pointed into the sun.
• Make him see people. Detection and tracking that holds up at a festival, in a driveway, and in a parking lot at night, because the cost of missing one is not symmetric with the cost of a false positive.
• Detect obstacles and hazards. What's in the path, how far away, how fast it's moving, and whether he should slow down or stop.
• Build toward driving without a driver. Chip Go! is remote-driven today, and your work is what lets him take more of it himself, one maneuver at a time.
• Watch him when nobody else is. Theft and tamper detection on a parked vehicle, held to a false-alarm rate a customer will actually tolerate rather than mute.
• Ship models to the Jetson. Quantization, latency budgets, and what that board can actually sustain in traffic on a hot day, not what the benchmark said on a workstation.
• Own the data loop. Capture from the fleet, labeling, evaluation sets, and knowing whether last week's change made anything better.
• Decide what not to automate. Some of this should keep a human in the loop for a long time, and you're the one who says which parts and when that changes.
• Work with the vehicle team. Drive-by-wire and networking are theirs. The seam between perception and actuation is yours together, and it's the one that matters most.

What you'll need

• Seven or more years building software, with real production computer vision in it.
• Perception shipped on a physical device, not only in a notebook. A robot, a vehicle, a drone, a camera product, something that had to work outdoors and keep working.
• Depth in modern detection and tracking. Training, fine-tuning, and the judgment to know when an off-the-shelf model is the right answer.
• Comfortable at the edge of the hardware. Latency budgets, quantization, accelerator quirks, and thermal limits on a board inside a hot vehicle. You don't need Jetson specifically, but you need to have shipped against a fixed compute budget and felt it push back.
• You've owned an evaluation loop. Datasets, metrics that mean something, and the discipline to believe them over a demo that looked great.
• Safety judgment. You think about what happens when the model is wrong before you think about what happens when it's right.
• Comfortable being the only person on this. No team under you at the start, though you can bring in contract help where it earns its cost.
• You write things down. What you tried, what it cost, and why the strange thing is strange.
• Based in Austin, Los Angeles, or Miami, or remote in the United States.

This is a wide brief and we know it. Perception, embedded, and safety thinking
rarely live in one person. Tell us which two you're strong in and which one
you'd be learning, and we'll take that seriously.

Bonus points

• Automotive or robotics autonomy stacks.
• Multi-camera calibration, or sensor fusion beyond cameras.
• Teleoperation or remote-driving systems.
• Jetson in particular, since that's what's on the vehicle. TensorRT, DeepStream, or CUDA on embedded.
• Rust.
• Low-speed, off-highway, or fleet vehicle experience.
• Video anomaly detection, or security and surveillance products.
• Simulation used for perception testing rather than demos.
• Published work in detection, tracking, or 3D perception.

What success looks like

• Chip stops for people reliably enough that we stop discussing it.
• A measured drop in how much of a Chip Go! trip needs a human.
• Theft and tamper alerts customers trust, which means rare and right rather than frequent and hedged.
• An evaluation loop that tells us whether a change helped before it reaches a vehicle.
• A second perception engineer who can be productive without reading your mind.

Compensation and benefits

• $160,000 to $200,000, depending on experience.
• Equity in Chip.
• Eligible for our company healthcare plan.