Harpoon
Log inBook a pilot
Human expertise for AI that builds in 3D

The 3D foundation for AI.

Every machine, robot, chip and building is designed in 3D first. As AI and automation learn to build the physical world, they need to get 3D right. Harpoon brings together real 3D experts from every field to train, test and fix the models that do it.

Get your 3D evaluatedJoin as a 3D expert
Or check a 3D model for free
Example · motor bracket from a text-to-CAD model
4 defects found
  1. 1Wall floats 2.5 mm above the base. The part isn't one solid.
  2. 2Mounting hole missing; another is undersized and off the grid.
  3. 3Rib cuts straight through the wall.
  4. 4Wall holes don't line up with each other.
Drag to turn the part.
Built forText-to-3DText-to-CADCoding modelsComputer-use agentsRobot simulationWorld models

AI can write an essay that reads well. Ask it for a bracket that fits, a robot cell that works or a floor plan that stands, and it still gets it wrong. The mistakes are small, technical and easy to miss. Only someone who does the job can catch them.

Fields

One foundation, every field that builds in 3D.

Each field has its own rules. We match every task with experts who work in that field.

Games, film & animation3D artists, riggers, animatorsTopology, UVs, rigs, facial shapes for lip-sync
Engineering & manufacturingMechanical engineers, CAD designersDimensions, tolerances, holes, can it be made
Robotics & physical AIRoboticists, simulation engineersJoints, collisions, mass, physics that holds
Automotive & aerospaceVehicle and aerospace engineersSurface quality, clearances, assemblies
Chips & electronicsPackaging and circuit-board engineersLayer stacks, heat paths, connections
Architecture & constructionArchitects, BIM specialistsStructure, building codes, model data
HealthcareMedical imaging and prosthetics specialistsAnatomy, fit, safe shapes
Retail & ARProduct visualization artistsReal-world scale, materials, how it looks
What labs get

Two ways we make your model better at 3D.

01RLHF

Expert evaluation

Send us what your model makes. Experts from the right field score each output, compare pairs, point out every defect and fix the ones that matter.

You get
  • Scores with written reasons
  • Preference pairs (A or B, and why)
  • Located defects and corrected files
02Datasets

Annotated 3D datasets

Millions of 3D models already exist, but they're messy and unlabeled. We filter the usable ones, check their licenses, annotate them and organize them into datasets ready to train on.

You get
  • Category, parts and real-world size
  • Quality score and defect list per model
  • Rig, materials and UV checks

Also available: expert-written 3D tasks, recorded expert sessions and tasks with an automatic check, for teams training coding models and agents. See all

How it works

A program first. Then an expert.

  1. 01

    Automatic check

    Every file is measured in seconds: holes, broken geometry, missing UVs, code that doesn't run.

  2. 02

    Expert review

    A professional from the right field looks at what a program can't judge, scores it and says why.

  3. 03

    Fix and deliver

    The outputs that matter are corrected. You get scores, comparisons and clean files.

Every expert is tested before they start and scored on hidden test questions while they work.

Task 0412 · Engineering · Compare06:41

Prompt: “L-shaped steel bracket, 80 × 50 mm base, four Ø8 mm mounting holes, two ribs.”

A
BPreferred
CriterionAB
Geometry is clean
Dimensions match the prompt
Can be manufactured
Why B is better
A is missing holes and its wall doesn't touch the base. B is one solid part and the holes sit on a clean grid. B's ribs are 2 mm too thin for steel.
Annotations

Every review becomes training data.

An expert doesn't just say “bad”. Each review is a structured record your team can train on directly.

Scores on a rubric
Each output is rated on the criteria that matter in its field: geometry, dimensions, rigging, physics.
Defects, located
Every problem gets a type, the exact place on the model, and how to fix it.
Preferences with reasons
Which of two outputs is better, and a written reason a model can learn from.
Corrected files
The expert's fixed version, so you have the before and the after.
Rig and face checks
Bones, skin weights and the facial shapes used for lip-sync (ARKit), checked by riggers.
annotation_0412.jsonexample
{
task: "0412",
field: "mechanical engineering",
output: "bracket_A.step",
scores: { geometry: 2, dimensions: 3, manufacturable: 1 },
defects: [
{ type: "not_one_solid", where: "wall / base joint", fix: "join wall to base" },
{ type: "missing_hole", where: "base, front right", fix: "add Ø8 mm hole" },
{ type: "intersecting_solids", where: "right rib", fix: "trim rib to wall face" }
],
preferred: "B",
reason: "A is missing holes and its wall doesn't touch the base.",
fixed_file: "bracket_A_fixed.step",
reviewer: { field: "mechanical", test_accuracy: 0.96 },
}

Example record from the review above. Delivered as JSON, with the fixed files.

Free tool

Check a 3D model for free.

Drop a .glb from any text-to-3D or image-to-3D tool. You get a score in seconds. No account, and the file never leaves your computer.

Drop a .glb file here
or click to choose one · stays on your computer
For AI labs

Start with a small pilot.

Send us a batch of outputs. You get back scores, expert reviews and fixes, and you decide if it's worth going further.

Book a pilot
For 3D experts

Get paid to teach AI your craft.

Artists, engineers, roboticists, architects and more. Remote, on your own hours, paid by the hour.

Apply as an expert
Our mission

Be the 3D foundation for AI: the place every lab and every industry turns to when AI has to understand, design and build the physical world.

The benchmark everyone cites

A public, expert-judged measure of how well AI handles 3D, field by field.

The first source of 3D data

Evaluations, annotations and original 3D data that AI labs trust and train on.

The largest 3D expert network

Artists, engineers, roboticists, architects and more, tested, scored and well paid.

The road ahead
  1. Now

    Evaluate and annotate

    • Expert evaluation of AI-made 3D (RLHF)
    • Annotated datasets from existing 3D collections
    • Games, film, animation and product 3D first
  2. Next

    Create and measure

    • Original 3D data, created by our experts for each lab
    • The Harpoon 3D Arena: a public ranking of AI models on 3D
    • Training data for agents that use 3D software
  3. Later

    Every field that builds in 3D

    • Engineering and CAD, robotics and simulation
    • Construction, automotive and aerospace
    • Chips and electronics, healthcare