Engineer, New Grad

Spectral Labs•San Francisco, CA

1d•$150,000 - $220,000•Onsite

About The Position

Spectral Labs is building foundation models for engineering physical systems. We are hiring new graduates for raw ability, not domain experience. What we look for instead of heavy experience with CAD kernels and generative geometry is evidence that you teach yourself hard things quickly and finish them. You’ll join our technical staff, and the role is deliberately broad. The work sits across machine learning, geometry, graphics, and AI: geometry and data pipelines, evaluation harnesses, internal tooling, and training and experiment support. We’re hiring a generalist, a real strength in at least one of those areas and genuine curiosity about the rest. You don’t come with a specialty yet, and that’s the point. A small team gives you the whole system, and you’ll find out where you’re strongest by working across it. You’ll get real problems early, and the people around you will help whenever you ask, but you’ll be the one steering. This is a fully in-person, full-time role at our office in SF.

Requirements

Graduating with a Bachelor’s or Master’s in Computer Science, Engineering, Math, Physics, or a related technical field, or graduated within roughly the last two years. You have a high GPA in a competitive undergrad program.
Real strength in at least one of machine learning, geometry, graphics, or AI, and genuine curiosity about the others.
Strong programming ability. You write Python and/or C/C++ comfortably.
Evidence that you finish hard things nobody assigned you: a substantial personal project, research you drove yourself, open-source contributions, competition results.
You love solving challenging problems.
Comfort with a high-ownership environment.

Nice To Haves

Objectively impressive achievements in any domain are a big plus. Built something unusual? Won something hard? We want to hear about it.
Research experience with a paper, preprint, or thesis you can talk about in depth.
Work on generative models: diffusion, autoregressive, or otherwise, where you trained something yourself and evaluated it honestly.
Experience generating synthetic data and showing it moved a real metric.
Experience building RL environments, reward functions, or post-training pipelines.
Coursework or projects in computer graphics or computational geometry: renderers, meshes, geometry processing, spatial data structures.
Any work modeling physical or engineered systems, simulation, control, robotics, motion planning, or physics-informed ML.
A track record of making things dramatically faster. Profiling, kernels, pipelines, latency: we care about the order of magnitude and how you found it.
Internship or contract experience where you shipped something real users depended on.
Competitive programming, math, or physics results (ICPC, IOI, IMO, Putnam, or similar).
You’ve read the papers in this space and have opinions about them.
Comfort with Linux, git, CUDA, and cloud infrastructure.

Responsibilities

Building and maintaining pieces of our geometry and data pipelines, parsing CAD, converting formats, extracting features, and verifying that what comes out matches what went in
Writing the evaluation harnesses and tooling that tell us whether a model actually got better
Supporting training and experiment runs: instrumenting them, debugging failures, and turning results into something the team can read at a glance
Building internal tools that remove friction for everyone else, the script nobody wanted to write that saves the team a day a week
Taking well-scoped pieces of larger research and engineering efforts and grow your scope as you earn it
Collecting data vital to our moat
Fixing things you notice are broken without waiting to be asked