SecureView

A secure gallery app built with Lynx
Encrypted images with PII
AI semantic search
Architecture

Problem statements

4 - Building UI for the AI Era with Lynx & 7 Privacy Meets AI: Building a Safer Digital Future

✨ Inspiration

In a world where every photo risks exposing identities, locations, or sensitive data, our app reimagines the photo gallery as a privacy-first sanctuary. Inspired by the needs of journalists, field workers, and compliance teams, it delivers instant on-device redaction, encryption, and auditability—ensuring that visual data remains secure from capture to sharing. This isn’t just a tool; it’s a shift in how we treat images as private assets, not public defaults.

🛡️ What it does

A secure gallery app built with Lynx, that leverages on-device AI for understanding what's on a image using VLMs and has powerful search interface to search images through natural language.

🛠️ How we built it

We leveraged Lynx for our cross platform user interface.
We used ONNX VLM models using the huggingface transformer library with vision capabilities to identify if an image has PII.
Images are run through an embeddings model to power semantic search.
Images with PII are encrypted with military grade AES-256 CBC encryption.

🚦Challenges we ran into

Using a new framework like Lynx was challenging. With limited online documentation and forums to rely on for troubleshooting, design patterns and component libraries. However, since Lynx was open source, we could dive into the source code to understand the implementations to come up with novel solutions to tackle the problem at hand
Finding a model that is capable enough to identify key features in an image, while fitting within the our self-imposed 1GB memory limit
Not much components other than very basic elements in building the application. So everything had to be built from scratch, from modals, routing to animations

🏆Accomplishments that we're proud of

Diving into the source code of a huge open source project head-first and that too within a few days of the hackathon announcement
Making the application work end-to-end, discovering novel AI/ML techniques
Getting a chance to collaborate with team members with different areas of expertise in crafting a unique solution while being within the problem statement

📚What we learned

The open source community is welcoming and provides lots of opportunities to learn and grow
While AI agents can work on established frameworks, they failed miserably in creating a Lynx application that would work as expected and had trouble sticking to the framework's elements even when provided with documentations and MCP servers.
Software Engineering, understanding and transferring learning from one tool to another is a human touch that can not be replaced - at least not yet!

🔮What's next for SecureView

The application currently works for web. We'd extend it more platforms like Android, iOS and even desktop operating systems.
Add features like clustering of people images by using face-embeddings
Support for custom Albums
Come up with a novel way to watermark images with PII to limit exposure and serve as access-control.