Every time you walk through a crowded tourist spot or attend a stadium concert, you become an extra in someone else's digital memory.
There is a massive, invisible digital footprint of your face scattered across the internet. There are likely thousands of photos of you sitting on servers right now that you will never know about. It begs a fascinating question: How cool would it be to find them?
Imagine being able to query the entire internet for your own face to find that random background shot of yourself in New York from 2018. But to build a tool that searches the open web for your face, a company has to relentlessly scrape billions of images without permission. It requires strip-mining personal data and violating privacy just to satisfy a curiosity.
I wanted to explore the technical side of facial matching without the shady data practices. That is why I built DopplGrid.
DopplGrid is a closed-loop, 100% private facial recognition network. Instead of scraping the open web, it operates as a secure vault. Our database only grows when real people explicitly choose to opt-in and lock their faces into the grid.
The Architecture Behind the Vault
To make this work securely, the app relies heavily on isolated backend microservices. The heavy lifting—like processing biometric vector embeddings and triggering real-time notifications—is handled securely via Firebase Cloud Functions.
For example, our onboarding flow utilizes a standalone, event-driven microservice. When a user joins the grid, a background function uses the Firebase Admin SDK to securely bypass frontend rules, provision a secure chat room, and inject a personalized welcome message along with real-time UI notification badges directly into the database. By keeping these services entirely isolated, the heavy backend processing never bottlenecks the user's onboarding flow and is completely abstracted away from the client side.
Because the network is entirely opt-in, you might not find a global doppelganger on day one. But the moment your face is in the vault, the engine is active. As more users join globally, your chances of an automated match increase.
Test the Engine
In the meantime, the matching engine is fully live. You can invite a friend, run your faces through the engine, and calculate your exact mathematical similarity percentage to see how closely you actually resemble each other.
We can build and explore the coolest parts of facial AI, but we have to do it on an architecture built on trust and consent.
Check out DopplGrid here and let me know what you think of the onboarding flow and the underlying architecture!
Note: Add the following tags in the Dev.to editor settings: webdev, programming, firebase, privacy
Top comments (0)