Datasetpublic

markov-ai/cad-environments

CAD Environments CAD Environments is a multimodal dataset of complete, human-performed workflows in desktop CAD software. The current release contains 51 task workflows totaling 99.03 hours, covering eight software groups across mechanical design, architecture, MEP, structural design, and general 3D modeling. Each workflow preserves the full task context—not just the final model—including the problem statement, reference and input files, a gold output, evaluation rubrics, a… See the full description on the dataset page: https://huggingface.co/datasets/markov-ai/cad-environments.

sourceHugging Faceupdated 2mo agoView on Hugging Face

17likes57kdownloads

Dataset Card

CAD Environments

CAD Environments is a multimodal dataset of complete, human-performed workflows in desktop CAD software. The current release contains 51 task workflows totaling 99.03 hours, covering eight software groups across mechanical design, architecture, MEP, structural design, and general 3D modeling.

Each workflow preserves the full task context—not just the final model—including the problem statement, reference and input files, a gold output, evaluation rubrics, a human walkthrough screen recording, synchronized interaction events, frame timing, and narration.

Dataset summary

Metric	Value
Tasks / workflows	51
Human walkthrough recording time	99.03 hours
Software groups	8

Software breakdown

Software	Tasks / workflows	Hours
AutoCAD	4	25.70
SketchUp	7	25.15
SOLIDWORKS	19	16.64
Revit Architecture	6	11.70
Revit MEP	1	7.19
Siemens NX	6	6.25
Revit Structure	3	4.26
STAAD.Pro	5	2.14
Total	51	99.03

What each workflow contains

—Problem statement: task_desc.json provides the structured task, application, domain, description, deliverables, and any specified dimensions.
—Human-readable task overview: task_overview.pdf packages the task for convenient review.
—Reference and source material: input_files/ contains the files and assets provided to the operator.
—Gold output: output_files/ contains the completed CAD deliverables produced during the walkthrough.
—Evaluation criteria: rubrics.json describes the requirements used to judge task completion.
—Human walkthrough: clip.mp4 records the complete on-screen execution of the task.
—Interaction trace: events.json contains synchronized mouse and keyboard events; frame_events.json provides frame-level timing and event alignment.
—Narration: narration.json provides timestamped descriptions of the workflow.
—Recording metadata: metadata.json includes duration, frame rate, frame count, event count, platform, and category.
—Derived computer-use representation: scale-cua-format/ contains trajectory annotations and step images for frame/action-based use.

Directory structure

data/workflows.parquet is a 51-row workflow index used by the Hugging Face Dataset Viewer. The raw contents of each workflow follow this structure:

text

<software>/<workflow-id>/
├── task_desc.json
├── task_overview.pdf
├── rubrics.json
├── metadata.json
├── clip.mp4
├── events.json
├── frame_events.json
├── narration.json
├── input_files/
├── output_files/
└── scale-cua-format/
    ├── annotations/trajectory.json
    └── data/images/

The canonical software prefixes are autocad/, nx-cad/, revit-architecture/, revit-mep/, revit-structure/, sketchup/, solidworks/, and staad-pro/.