Hi Streamlit Community! ![]()
I recently launched Filigrane Malin, a 100% local-first and zero-trust application (available as a desktop app and Docker container) designed to securely protect and watermark identity documents (ID cards, passports, etc.).
The Problem: Uploading unencrypted, highly sensitive documents to centralized remote servers (even official government ones) creates a massive “honeypot” for hackers. I wanted to build an alternative that proves regulatory compliance (GDPR) is not the same as maximum security.
The Streamlit Twist (Zero-Trust Demo): I wanted to share a live web demo of the watermarking engine, but doing so natively violated the app’s core Zero-Trust philosophy. If I included a standard file uploader, users might accidentally upload real, sensitive documents to the cloud.
To solve this, I built a dedicated demonstration on Streamlit Cloud with a privacy-first architecture:
-
Zero File Uploads: I physically removed
st.file_uploader. -
Pre-Loaded Specimen: The app uses a pre-loaded official French national ID specimen, allowing users to safely stress-test the engine.
-
Full Feature Set: Users can live-test the sine-wave text distortion, alternating drop shadows, and the CNIL-compliant password generator (using Python’s
secretsmodule) and strength estimator.
Tech Stack: Python 3.13, Streamlit, Pillow, PyMuPDF.
Links & Resources:
Live Streamlit Community Cloud Demo
Source Code, Release & Container on GitHub
Project webpage
Read the full behind-the-scenes story on Medium
I would love to hear your feedback on the app and this approach to building privacy-first demos in Streamlit! ![]()