Skip to content

Training Platform

Status: Active supporting infrastructure
Last synthesized: April 2026
Purpose: Home for the labeling, detector-training, and runtime experimentation surfaces that support drawing-analysis work.

What This Section Is For

Training Platform covers the practical infrastructure behind drawing-analysis model work: how data is labeled, where detector experiments live, how models are trained, and which local or hosted runtime paths are currently good enough to reuse.

This section sits underneath Infrastructure because these training and runtime systems enable RapidDraft and CAD-intelligence work, but they are not themselves the product surface.

Open First

Current vs Reference

  • Current: the active operational reality now includes CVAT on Fedora, Nuclio-linked model serving, Vast.ai training, W&B dashboards, and parallel YOLO plus Roboflow experiment paths
  • Current: the DraftLint repo is still useful reference history, but the most important active workflow now lives in the dedicated training-operations material
  • Reference: older tool-survey material still matters, but the current operational reality now lives more in the repo/runtime assessment pages

Section Map

Sources