Linda Krestanova4 min

Edge AI: Save Money, Protect Users & Beat Competition

EngineeringProductOct 18, 2022



Oct 18, 2022

Linda KrestanovaCommunications Manager

Share this article

Edge AI is a ton of ML algorithms that run on a physical device, where content doesn’t have to leave that device. While that’s a hard concept to grasp, the business benefits are as straightforward as it gets.

Before we get into those benefits and how they impact your business, let’s start with an example.

Our client, Zooni, wanted to build a creator social media platform with user-generated video content. A part of that was a “Discover” video feed for that scrolling we all love — Gen Z most of all. And Gen Z includes a lot of kids, so inappropriate content has to be filtered out, FAST.

Two challenges presented themselves:

  1. How do we flag inappropriate content at record speed without breaking the bank?
  2. How do we do it for video and live streaming (a lot harder to track than photos)?

The only answer to both of these questions is Edge AI, specifically on-device content moderation. At the moment, there’s no competition. You cannot guarantee security, privacy, speed and affordability with anything else out there.

Understand Edge AI in a Few Sentences

Typically, AI runs on a server somewhere in the cloud. With Edge AI, it runs directly on the device, like your mobile phone. This way, it works offline and quickly, since it doesn’t have to communicate with the server. (More on the benefits below.)

Take autonomous car navigation systems. While AI is trained in the cloud, everything happening while driving — controlled steering, acceleration and braking — is run directly in the car.

Why Is Having an AI/ML Solution “Close to the User” So Great?

Running the solution on a user’s physical device means users get data in real time — no other systems or internet connection needed — and a user’s content doesn’t have to leave the device.

The benefits transform your product and the user experience in multiple ways.

  • Increased security and privacy — No sensitive photos of the user are sent over the internet
  • Speed, speed, speed — No connections between the user’s phone and a server happening in the background
  • Reduced costs (by a lot) — No expensive infrastructure in a cloud; the solution scales on its own with each new device
  • Better management and utilization of resources — No need to transfer content into cloud storage just to get it validated and then have the result returned back — which is great, especially for multimedia content
  • Personalization of your product feels more natural — Running ML on-device lets you retrain the model and personalize it to your preference (think smart keyboards with your dictionary of words); in case of content moderation, rejected false positives are minimized

Back to Our Example: Saving Money, Outperforming Competition

Zooni trusted us in part because STRV is one of the few software development companies offering an on-device content moderation solution comparable to Facebook’s or Apple’s.

We’re able to develop, train and integrate a custom content moderation model smaller than 10MB which can be embedded into any mobile or web app.

  • As soon as we integrate the model into your app, content moderation doesn’t cost you anything. For a Twitch-sized company, savings can reach 99.99%.
  • Our solution scales naturally with each device because it occurs on the user’s device. For Zooni’s video content, our model scales based on video length, quality and quantity.
  • Sensitive content isn’t sent to a third-party service, which negates security threats. The user’s data is processed on his/her own personal device; all content stays with the user.
  • Because there’s no need to upload large files to third-party servers, our model comes with instant processing speed. Results are handled immediately on the device.

An Engineer’s Perspective: Edge AI in Mobile App Development

When incorporating Edge AI into the mobile app development process, a data science engineer faces a few challenges. Nothing detrimental to the process, timeline or wallet — as long as the engineer’s already been around the Edge AI block a few times.

Integration/deployment, managing resources and monitoring

Making sure that the solution fits on the device means taking extra steps, doing a lot of optimization and keeping an eye on shifts in data; failing to do so may leave you with a useless solution.

Sourcing relevant data and gaining domain knowledge

More often than not, the necessary data infrastructure isn’t in place at the beginning of the process. It’s crucial to work hard (and smart) to gather a reasonable amount of data for the problem being solved.

Carefully introducing assumptions and biases

Basing decisions on assumptions requires having sufficient domain knowledge. When done thoughtfully, it can significantly decrease the data needs.

Building the solution iteratively

It’s important to be evaluating and improving on the latest results of the solution constantly, to ensure that you’re solving the right thing with the appropriate solution.

Training the model (when done on device)

Integrating on a mobile phone comes with possible inconsistencies with certain legacy models — for instance, users might have an old model so there may be compatibility issues.

In general, to make the solution a functioning part of the app, working very closely on integration with the mobile engineer is a must.

Should You Consider Using Edge AI?

Probably. There are countless Edge AI applications. To name a few: personalized content, facial recognition, wearable devices, smart gadgets and speakers, semi-autonomous cars, computer games, robotics, drones and surveillance cameras.

We’re aware that this is not the easiest topic and are always open to having a conversation about the possibilities, no strings attached. So if you have any questions, we’re happy to chat.

Share this article

Sign up to our newsletter

Monthly updates, real stuff, our views. No BS.