Machine Vision AI Cuts Moderation Time from Hours to Seconds

Smooth Moderator: Adrennial’s Client Success Story

  1. Powerful Machine Vision AI keeps content PG for adventure media streaming company Adrennial, cutting moderation time from hours to seconds.
  2. Ultrasecure hybrid-cloud architecture scales AI using Kubernetes to manage cost without sacrificing performance.
  3. Hyperefficient data pipelines in market leading Google Cloud infrastructure build a data warehouse to identify content-policy breaching users.
  4. Brand partners feel secure, users feel safe.

    With Distributed Analytics we unlocked confidence and time.

    Adrennial’s CTO

Meet Adrennial

Adrennial is the premier social video app for those interested in adventure, travel, outdoor living, and alternative lifestyles. Established in 2019, their video hosting platform is a fantastic way for cutting-edge producers to share adventure content. The company now showcases media from 200 interest verticals to a global audience – offering an incredible reach for brands targeting a progressive, proactive base. In return for audience access, brands can help creators monetise their work through partnerships and advertising: in turn enabling them to deliver the best possible content without distractions.

Everything in Moderation: Media Policy for Client Success

Of course, partner brands want to know that the content on which they advertise meets certain standards of decency. Similarly, Adrennial themselves must ensure that content they host falls within the limits of ‘adventure’ content, and the law. How can this be achieved? Enter – the moderator.

Content moderators, or mods, are an essential part of the online ecosystem. The moderator is the first, and often only, line of defence against rogue content. Moderators must identify, then shut down, inappropriate uploads before things get out of control. But – mod powers are limited. There are only so many hours in the day, and human mods must spend most of them watching content to search for objectionable inclusions. To enable this, hosts must either place newly uploaded content in a long backlog hidden from view until the moderator can get around to assessing it, employ a huge moderator team at high cost, or allow uploads to be shown to users straight away and rely on user-reporting to flag inappropriate content for mod review. None of these options are ideal, and in all cases human error means a 100% success rate for blocking rogue content is next to impossible to achieve.

Are You Afraid of the Dark?

The video portal was reliant on exactly this kind of human moderation during its pilot stage. A small mod group drawn from the development team manually reviewed all uploaded content to check for inappropriate inclusions – even during this scoping period, multiple hours were being spent daily assessing content suitability.

We were feeling despondent, we didn’t know how we were going to make this work on the scale we needed. It was like being trapped in the dark.

Adrennial’s CTO

This pulled staff away from other key responsibilities, and if deployed during standard site operation would have halved the development team or incurred huge costs paying moderators. An automated system was needed. The company looked at several existing automoderation tools; finding them unsatisfactory due to their closed black box nature, high costs, and obfuscation of results – delivering metrics for efficacy that the company could not verify. Adrennial’s CTO said “we were feeling despondent, we didn’t know how we were going to make this work on the scale we needed. It was like being trapped in the dark. It was unsettling. We’d done all this work and felt trapped at the final hurdle”.

Turning On the Lights

That’s when the company’s CTO found Distributed Analytics. Through a quick consultation, Distributed Analytics were able to identify key challenges and propose a solution leveraging Artificial Intelligence experts on our engineering team.  “Distributed Analytics just switched the lights on” said the CTO “they listened to our problems and had a solution on the table almost immediately”.

Distributed Analytics just switched the lights on… they listened to our problems and had the perfect solution on the table almost immediately.

Adrennial’s CTO

Distributed Analytics’ AI engineers crafted a custom, lightweight, cost-efficient Machine Vision GAN AI able to assess the video content uploaded to Adrennial’s platform – an automated moderator, or ‘automod’. This Artificial Intelligence was hosted on ultrasecure hybrid-cloud infrastructure established using Antos, offering unparalleled scalability through use of Kubernetes and Google Cloud Functions.

Flexible scaling offered Adrennial the performance they needed for a cost they couldn’t believe. “I was floored” reported the CTO, “the automod was surpassing the speed and success of our human moderators at a price that annihilated the generic solutions from other vendors”. The company’s CTO went on to state that “total human hours were reduced by roughly 80%… and review times for even the longest and most complex videos fell by 45%, we were thrilled”. Moderation time was slashed from hours to seconds, and nothing slipped through the automod’s defences – helping easily maintain content standards.

Total human hours were reduced by roughly 80%… and review times for even the longest and most complex videos fell by 45%, we were thrilled.

Adrennial’s CTO

This represented a major success: second-scale moderation for shorter content, strict cost control, and 100% success in blocking objectionable content. But the power of the ‘automod’ did not stop there, also unlocking benefits for the client company in terms of greater monetisability and smooth content management.

Levelling Up

Custom APIs allowed the Distributed Analytics ‘automod’ to flag and block content delivery through interacting with Adrennial’s chosen streaming solution – Wowza. Distributed Analytics data experts then implemented a hyperefficient Google Cloud Functions pipeline. This pipeline ingested content policy violation data through Pub/Sub, then distributed it via Dataflow into Firebase databases warehoused in BigQuery. This dual blocking-logging approach prevented inappropriate content being seen by users, pleasing brand partners, but also provisioned a database – allowing Adrennial to identify uploaders regularly breaching content policy and act accordingly.

It was just the right solution… with Distributed Analytics we unlocked confidence and time

Adrennial’s CTO

Additionally, through utilising Google’s leading cloud data management technologies, Distributed Analytics were able to offer serverless data warehousing affording scalable analysis for petabytes of data – providing a futureproof data solution that grows alongside the client.

For the CTO “it was just the right solution,”. Content could be moderated and he “could literally see it working by looking in the BigQuery warehouse”. Brand partners were thrilled, users were happy, and everything was working faster – “with Distributed Analytics we unlocked confidence and time”, meaning the team “could focus on building brand relations, showing potential partners almost anything on the platform and feeling secure that it was never going to be something that breached content policy”.

Client Success with Distributed Analytics

The ‘automod’ GAN AI blocked 100% of policy-violating content once deployed. By hosting this solution on ultrasecure hybrid-cloud infrastructure combining private servers and Google Cloud, Distributed Analytics were able to offer a scalable, cost-efficient solution that slashed human moderation hours by 80% and reduced review times for individual videos by at least 45% – even in the most extreme cases.

In media, hosting objectionable content can affect reputation, brand partner trust, and user experience. Adrennial needed a bulletproof solution for a competitive price, and that’s why they chose Distributed Analytics.

Join the conversation by commenting below! Or click here to discover more Distributed Analytics client success stories

Leave a Reply

Your email address will not be published. Required fields are marked *