LABCRAFT

Features

One vision engine, two places to use it.

Inspect runs on production lines. Classroom runs in a browser for K-12 students. Both share the same labeling, training and evaluation lab.

LabCraft Inspect

Defect detection and video analysis for production lines

  • Known defect classification. Train on labeled scratches, dents, missing parts or misprints and sort every frame into a class.
  • Anomaly detection. A separate model learns what a good part looks like and flags anything unfamiliar, so new defect types are not missed.
  • Object detection and counting. Locate and count parts, components or packages in each frame.
  • Video stream analysis. Run models on continuous camera streams, not just still images.
  • Event log. Every result becomes a searchable event, and flagged frames go to a review queue that feeds new training data back in.
Illustrative product interface

Training lab

Label, train, evaluate, export

The experiment environment behind both products. Compare training runs, tune thresholds against held-out images, and export models for the hardware you run.

Label types

  • Bounding box
  • Polygon
  • Keypoint
  • Oriented box (OBB)
  • Instance segmentation
  • Semantic segmentation
  • Classification

Export formats

ONNX and TensorRT, plus a hosted API endpoint.

inference result (example)

{
  "frame": "F-20420",
  "line": "line-2",
  "model": "bracket-defects-v7",
  "detections": [
    { "class": "scratch", "confidence": 0.91, "bbox": [412, 188, 96, 34] }
  ],
  "anomaly": { "score": 0.64, "status": "review" },
  "timestamp": "2026-11-04T10:42:08Z"
}

Illustrative response format.

LabCraft Classroom

A 3D virtual lab where students train image models

  • Interactive 3D scenes. Students move a virtual camera around objects, change lighting and angles, and capture their own training images.
  • Train and test in class time. Label, train a small model and test it on new scenes, then see which examples it gets wrong and why.
  • Built around AI4K12. Units map to the AI4K12 Five Big Ideas, starting with Perception: how computers sense the world.
  • Teacher controls. Class rosters, assignment of units, and a view of each student's runs.
  • Browser-based. Runs in a web browser, with no student photos required.
Illustrative product interface

Deployment options

Cloud API

Send images or streams to a hosted endpoint. The fastest way to start a trial.

Edge devices

Run exported models next to the camera on devices such as NVIDIA Jetson, with results published over MQTT.

On-premises

Offline installation for plants that keep images inside their network. Enterprise plans.

Security and privacy

Data ownership
Factory images and models belong to the customer. They are not used to train other customers' models.
Student privacy
No advertising, no sale of student data, no profiling. Deletion at the school's request. District DPAs supported.
Data minimization
Classroom labs default to virtual-scene images, so no student faces are needed.
Access control
Role-based access for teachers, students, engineers and admins. SSO on Enterprise plans.
Model licensing
We tell you the license of every model architecture you train, so commercial use is clear before deployment.

Getting started

  1. 01

    Send us a sample

    A few dozen images of good and defective parts, or your grade level and device setup.

  2. 02

    Set up a trial

    We create your workspace: a line in Inspect or a school in Classroom.

  3. 03

    Label and train

    Label in the browser, run training and check results against held-out images.

  4. 04

    Deploy or teach

    Connect the model to your camera stream, or assign the lab units to a class.

Start a trial See pricing