Product builders around the world frequently trade-off between three strategies<\/p>\n\n\n\n
When should you (re)-build your models and invest in iteration to get great quality? When should you borrow high-performance APIs and free your developers to make the next big thing? And when should you buy solutions for your problem?<\/p>\n\n\n\n
This week, we talked to VideoKen<\/a> co-founder & CEO Vishnu Raned<\/a> and his team, to understand how they traded-off between \u201cbuild\u201d and \u201cborrow.\u201d<\/p>\n\n\n\n VideoKen is an AI-powered video interactivity solution that transforms learning videos into interactive and immersive experiences. In this age, attention is at a premium, and learners have access to more information than they can consume. \u201cVideoKen\u2019s purpose is to take learner engagement to the next level by making it easy for people to discover and interact with videos<\/strong>,\u201d says Vishnu. VideoKen offers automatic video chaptering, in-video quizzes, deep video search, video analytics, closed captioning, video hosting and other features to make this interaction rich and impactful.<\/p>\n\n\n\n VideoKen\u2019s solution uses AI to track, predict and improve learner engagement using in-video interactivity and actionable insights. The goal is to minimize the problem of poor engagement among learners.<\/p>\n\n\n\n Over 100+ learning leaders leverage VideoKen to help learners engage and interact with videos using an auto-generated navigable storyline, deep video search and in-video assessments<\/strong>. These leaders also tap into VideoKen\u2019s in-depth analytics to gain insights into their learners\u2019 preferences and behavior, which is used to optimize and prioritize their learning content.<\/p>\n\n\n\n \u201cVideoKen\u2019s customers have been successful by driving 3x more engagement and retention on their learning videos,\u201d according to Vishnu.<\/p>\n\n\n\n VideoKen\u2019s journey started by developing in-house technology (now patented) for improving engagement. On top of their five patents, VideoKen needed to be able to offer a comprehensive solution to meet growing customer needs. \u201cThe engineers were eager to build in-house models, but it turned out we couldn\u2019t do it all by ourselves,\u201d says Vishnu.<\/p>\n\n\n\n Moreover, VideoKen needed an adequate cloud computing infrastructure which led to exploring and experimenting with various options before settling on Azure.<\/p>\n\n\n\n Jeril Sebastian, Head of Engineering at VideoKen, shared insights on using some of the applied AI services. \u201cMost of our services rely on VideoKen\u2019s own AI technology. One thing that we do rely on externally, is Azure\u2019s Cognitive Services<\/a> for transcribing videos.\u201d<\/p>\n\n\n\n Jeril explained that before leveraging Microsoft Azure and its Cognitive Services, VideoKen was using a combination of its proprietary speech engine along with some services of another cloud provider. \u201cIt was taking up a lot of time and wasn\u2019t feasible for us to pursue this approach long-term,\u201d Jeril said, adding that the company was spending twice as much money on the in-house model compared to Azure\u2019s speech API. \u201cEven though everything was working well with our AI speech-to-text solution, we needed to focus on our key strengths and have our developers work on solutions that are important to our customers<\/strong>.\u201d<\/p>\n\n\n\n After discovering that hosting GPU instances and training models in-house was taking up too many resources, VideoKen decided to test different options, with Azure\u2019s speech to text<\/a> proving a perfect fit. \u201cSpeech to text is a single building block of a much bigger infrastructure, so we couldn\u2019t afford to lose a lot of time and energy, despite being satisfied with the overall performance of our home-grown technology. Therefore, replacing it with Azure was the most logical step for us.\u201d<\/p>\n\n\n\n Video transcription has become a norm nowadays. One of the most important aspects of making videos more immersive is ensuring that learners do not miss any important information. Having realized Azure Cognitive Services\u2019 power, Jeril and the engineering team managed to cut the cost of video transcription by 50%.<\/p>\n\n\n\n \u201cWe used to have an ML engineer just to train the model, improve it, run the GPU instances, and so on,\u201d Jeril said. \u201cThis was a full-time job for one person, and now it takes just a few hours a week for me to take care of that entire pipeline. Our ML engineer now has more time to focus on researching and upgrading other services offered by VideoKen, instead of doing the grunt work to keep our in-house transcription model competitive<\/strong>.\u201d<\/p>\n\n\n\nWhat is VideoKen?<\/h2>\n\n\n\n
What happens behind the scenes at VideoKen?<\/h2>\n\n\n\n
VideoKen\u2019s development journey<\/h2>\n\n\n\n
Developing in-house machine learning models for speech-to-text transcription<\/h2>\n\n\n\n
VideoKen\u2019s journey post-Applied AI<\/h2>\n\n\n\n