---
title: ShockVortexML Interactive Demo
emoji: 🌪️
colorFrom: blue
colorTo: purple
sdk: static
app_file: index.html
pinned: false
license: mit
short_description: Neural shock-vortex segmentation in compressible CFD
---

# ShockVortexML Interactive Demo

Explore the multistructure pipeline used in published Movies S9 and S10.
The left panel shows density schlieren; the right shows the learned local shock
front/band, retained core/wake outputs and PM-v4 expanding-flow region.
The figures use saved movie predictions, with provenance in
`demo_provenance_v2.json`. Blue denotes expanding flow, not a validated centred
Prandtl–Meyer fan. These are previously inspected development examples.

The three HJ checkpoints and quantitative control results in the linked model
repository are a separate task-preservation experiment. They do not by
themselves reproduce this extended movie pipeline. The free Static Space serves
precomputed examples and links to the source and full-resolution movies.

This is a research demonstration, not human ground truth or a universal CFD
diagnostic. See the linked model card and
[GitHub repository](https://github.com/Ehsan-Roohi/ShockVortexML) for the exact
input contract, limitations, hashes and reproducibility material. Original
supplementary movies are available in the
[GitHub media release](https://github.com/Ehsan-Roohi/ShockVortexML/releases/tag/movies-localfront-v2-20260909)
and on the [author’s YouTube channel](https://www.youtube.com/@ehsan_roohi).