Kosher Vision
On-device image filtering with religious sensitivity — offline by design.
Overview
Kosher Vision is an on-device TensorFlow-Lite model that intercepts WebView images and hides non-tznius content before the pixel reaches the user. Four-level classification (NSFW → religious modesty → violence → symbolic). Embedded in the browser APK, ~5–15 MB quantised. Never leaves the phone. v0 POC in progress, deadline Q2 2026.
What it does.
Active construction. Scope and architecture locked; implementation in progress.
Where this is heading.
v0 POC with off-the-shelf NSFW dataset (deadline Q2 2026), v1 with a custom-trained religious-modesty classifier, integration into Kosher Browser as a WebView image interceptor. Stays on-device throughout.
Closed beta — request access.
This product is gated behind closed beta. Leave your email — every request is reviewed personally and access is granted in batches as testing slots open.