PrivacyLens

Mock-up For You Page
Upload page for creators, able to upload from gallery or take a new video
Video editing page, before PII is flagged out with standard editing tools
PII flagged out and option given to user to blur detection
Final blurred outcome with object detection

Inspiration

Millions of videos hit TikTok every day. People share their lives, run businesses, and build audiences with nothing more than a phone camera. However, there is a growing problem. Private information slips into the frame without anyone noticing. A casual vlog shows a credit card on the table. A live stream reveals a home address in the background. These accidents can lead to stolen identities, lost trust, and real financial harm. Most creators never realize it until someone else points it out. Seeing this gap inspired us to build a better way to protect people before mistakes happen.

What it does

PrivacyLens is a mobile-first video protection tool.

Users record or pick a video in the app.
While they edit, our backend scans every frame using YOLO detection to flag sensitive data.
The UI shows flagged frames for review.
Users can upload the original or choose a blurred version where flagged regions are automatically redacted.

How we built it

We built PrivacyLens as a cross-platform mobile application using React Native and Expo, allowing users to record or upload videos and interact with a familiar social media–style interface. On the backend, we ran a FastAPI server locally and expose it via an ngrok HTTPS tunnel that manages video uploads, frame extraction, and detection tasks. Our detection pipeline integrates custom-trained YOLO models for credit card and license plate detection, combined with OCR and validation checks for higher accuracy. Once detections are complete, the backend returns flagged frames to the frontend, enabling users to review, approve, or discard them before deciding whether to upload the original or a protected, blurred version. The system runs asynchronously so video analysis happens in parallel while users edit, ensuring minimal disruption to the overall experience.

Challenges we ran into

Building a real-time video privacy tool brought several challenges. Handling large video files on mobile devices while maintaining smooth interaction required careful API design and efficient encoding strategies. Running detection on every frame without slowing down the workflow pushed us to explore batching techniques, GPU acceleration, and pipeline parallelism. We also had to strike the right balance between detection sensitivity and false positives so users would not be overwhelmed with unnecessary alerts. Video rendering was another hurdle since applying blur to sensitive areas while preserving video quality and audio synchronization required extensive testing and optimization.

Accomplishments that we're proud of

We are proud that we built a fully working end-to-end pipeline within the hackathon timeframe. From the mobile interface to the detection backend to the rendering pipeline, every piece fits together smoothly. The real-time detection that runs in parallel with editing gives the app a natural user experience, and the modular backend design means we can add new types of sensitive information detection without rewriting the system. Deploying the entire solution on Render also makes it accessible to anyone right away.

What we learned

Through PrivacyLens, we learned how to integrate computer vision, OCR, and real-time video processing into a mobile application without compromising user experience. We gained practical experience in deploying FastAPI services on cloud platforms, handling asynchronous workflows, and optimizing media pipelines for speed and reliability. Most importantly, we saw the value of building privacy protection tools that work quietly in the background, giving users confidence and control without adding friction to their creative process.

What's next for PrivacyLens

We plan to take PrivacyLens far beyond its current hackathon version. A major priority is enabling on-device detection to improve privacy, eliminate dependency on internet speed, and significantly reduce latency. We also aim to broaden detection capabilities to include more sensitive information such as passports, license plates, and faces. Future updates will give users finer control with manual tools to adjust or remove detection boxes before applying blurring. We plan to integrate scalable cloud storage solutions to handle larger uploads efficiently and introduce batch video scanning features designed for professional creators managing high volumes of content.