Research Engineer, Design to Manufacturing

San Francisco · in person · $250k-$450k + equity

Applying here considers you for research roles across Vizcom. Roles are defined by the person, not the posting.

Vizcom is where design teams at Nike, GM, New Balance, and Hasbro do their daily work: sketch, render, color and material, 3D, export. The render was never the point. The point is the physical thing. We call it pencil to product. We're a Series B company with over $52M raised from Radical Ventures, Index Ventures, Unusual Ventures, and Basis Set.

Visualization is nearly solved. Making is not. Between a beautiful render and a part a factory accepts sits a canyon: meshes that aren't solid, surfaces that can't be edited, geometry that ignores how things are actually built, and softgoods that render perfectly but can't yet be flattened into patterns, sewn, or knit. Our customers design shoes, cars, power tools, and toys, and their asks arrive weekly: an engineering team wants generated 3D granular enough to hand to mechanical engineers. A yacht studio hand-traces curves from generated orthographic views into NURBS software. Color teams need thirty colorways on one hero product where the shape cannot drift at all.

No single field answers them. A render says what something looks like. A factory needs what it is: form, material, dimension, behavior, and the documents that carry them. Geometry is the spine of that chain, but only one link in it. The whole chain is where image generation was five years ago: representations unsettled, benchmarks missing, the winning path unknown. Our slogan is make it real. This role is the literal version of that sentence.

The role

As a research engineer here, you'll close the distance between pixels and atoms. You'll own the visualization-to-actualization chain as an open research direction with a product attached: which links to attack first, which representations to bet on, and what factory-acceptable means, written down and measured. Geometry is likely where the work starts. It is not where it ends.

You won't work alone. A preference-data engineer is capturing how designers actually work, and the post-training engineers are modeling what they choose. You make what they choose physically real, and eventually the loop closes: manufacturability becomes a signal models can learn from, and designer intent becomes a constraint on geometry. If you want to make prettier pictures, this isn't it. If you want a factory to accept a file that started as a sketch, it is.

We don't have a religion about representations. Meshes, splats, implicit fields, parametric surfaces, CAD operation sequences: whichever survives contact with a factory wins. The field hasn't picked a winner, and neither have we. What we have is the half of the problem nobody else owns: the designer's intent, captured from the first sketch, at enterprise scale.

What you'll own

• The map of the chain: from render to real, which links are attackable now and which are bets for later.

• Form: generated 3D to editable, manufacturable geometry. Parametric surfaces, curve extraction, CAD handoff.

• Material and color: rendered appearance to real-world specification. A finish that exists, a color that survives daylight, a textile with a name.

• Softgoods: pattern flattening, seam logic, knit. The half of the world that gets sewn instead of molded.

• Manufacturability in the loop: manifoldness, tolerances, wall thickness, drape and fit as properties of generation, not a repair pass after it.

• The output layer: the artifacts downstream teams actually consume (CAD files, pattern pieces, tech packs) and the eval bar for factory-acceptable, defined per vertical.

• Customer pilots and the flywheel bridge: real weekly asks as your test cases, and with the preference and post-training roles, turning constraints and judgment into signals models learn from.

This list is a charter, not a week one to-do. Nobody runs all of it at once, and the sequencing is yours to argue for.

First 90 days, one way it could go

• Days 1 to 30: immerse. Sit with the customer asks, the current 3D pipeline, and the failure gallery. Talk to the designers, and to the engineers downstream of them.

• Days 30 to 60: demonstrate. Take one real customer workflow and remove one manual step from it, end to end.

• Days 60 to 90: chart. Write the actualization roadmap: which links of the chain, in what order, and how we'll know they're working.

We expect you to

• Be deep in at least one link of the chain (geometry processing, appearance and material modeling, cloth or textile simulation, computational fabrication) with working fluency across the others.

• Have hands-on experience with modern generative or reconstruction methods, and the engineering skill to turn papers into pipelines.

• Be comfortable being first. There's no research log to inherit here. You'd be starting it.

Nice to have

• CAD kernel, B-rep, solid modeling, or class-A surfacing exposure.

• Color science, appearance capture, or knit and textile engineering.

• Physical simulation that runs at product speed: cloth, FEA, or structural sanity checks.

• You've shipped a physical product, or watched one die in the handoff.

And the disposition we keep coming back to: you judge a generated model by whether it could exist, not by how it looks spinning.

What you get

• A frontier that just opened: everyone is training image models, and the race from generated design to production has barely started. You'd run it from the design side, which nobody else owns.

• Live demand: enterprise design teams asking for this weekly, and design-challenge winners already heading to physical production as your test cases.

• The team: a preference-data engineer capturing how designers work and a post-training engineer modeling what they choose. Your constraints become their reward signals.

• A CEO who trained as a transportation designer at Honda and has personally lived the handoff you're fixing. For this role that isn't a fun fact. It's peer review.

How we work

• Research-log culture: what we learned, not what we worked on. Negative results are celebrated.

• Nothing scales on vibes. A result repeats before it earns compute.

• We publish what we learn: writeups and showcases, negative results included.

• We hire through paid work trials on real problems with real data, not LeetCode.