WEBVTT

00:00:00.000 --> 00:00:10.000
The model you trained. Still working for you. IP Vault seals selected app assets and controls installation access.

00:00:10.000 --> 00:00:20.000
Our reference model learns handwritten digits from generated strokes. The baseline APK hands over an ordinary, reusable model file.

00:00:20.000 --> 00:00:40.000
The model is sealed. The app still predicts. Recorded stroke inputs go into a real model session. The displayed predictions are computed by the app.

00:00:40.000 --> 00:00:52.000
20 reference drawings. Matching predictions. Both minified builds match all 20 reference inputs. The same asset searches catch the baseline and find no declared plaintext probes in the protected package.

00:00:52.000 --> 00:01:04.000
Investigating a running app takes more work than unpacking a file. The active model is still readable in process memory. This reference model does not establish general handwriting accuracy.

00:01:04.000 --> 00:01:12.000
Run a free local scan at ip.rev3rse.dev/scan, then scope one asset. Android supported; iOS coming soon.