HypeHop wanted to pay real cash to users who watch videos online. To do this, it had to have a way to watch users. Visible Magic created a system that let users watch the videos and rewarded their time and attention.

The solution was complex: because users were wary, the team wanted a secure way to confirm users were watching their videos without invading user privacy. The functional requirements also stated that the system should be able to use machine learning algorithms to recognize emotion and disinterest by watching an anonymized model of the user’s face.

Because the technology was potentially invasive, we also had to be very careful to ensure absolute user anonymity. Because the users were paid in cryptocurrency and their faces were never stored on the backend, we could promise users that their wallets could never be cross-correlated with their faces. That said, the stakes were high: these kinds of tools were difficult to build and expensive to maintain. They were taking a flyer here and they wanted something under budget in order to test their hypothesis.

Visible Magic launched the product in four weeks. The first two weeks were spent creating the complex, blockchain-based backend and front-end facial recognition system. The rest of the time was spent adding features and finalizing the interface.

The front end was a React application connected to a customized video player that connected to a computer vision system. The backend used a Node.js for the blockchain interaction and the team then used a number of pre-computed face-tracking models to add the eye tracking functionality. The final part, the emotion sensing, took a few days.

The app used many ready-made AWS services as well as custom code written for Visible Magic’s stable of crypto projects.

After launch, the team realized that fraud was a real concern. Users would watch the same video over and over again just to gather free cryptocurrency. To fix this, Visible Magic added a biometric and IP-based tracking system that would flag users who might be abusing the system.

Because the app was supposed to work on both desktop and mobile, the team began testing the computer vision model on a number of phones.

“We faced a lot of pain in working with a mobile browser because it was really hard for mobile phones to track computer vision in real time while playing a video and buffering the stream,” said Visible Magic CTO Viktor Shpak. “It was quite complex and our mobile phones were burning up during testing because they were doing so many things at once. After a bit of coding, however, we had a reactive app that worked on the desktop and the phone.”

Once Visible Magic worked out the bugs, HypeHop could prevent repeat offenders from stealing too many views and the system even allowed them to embed the video player on other pages, a feature that was a “nice to have” and became surprisingly popular.

The final product was a success. Visible Magic and HypeHop teamed up to build something highly unique and the company was pleased with the results. “Visible Magic makes the hard things easy,” one client rep told us during a final call.