The First Artificial Intelligence Framework for the Video Streaming Industry
Jump Data-Driven intelligence created HAMLET, the First AI (Artificial Intelligence) Framework for the Video Streaming Industry in order to maintain its position of leadership in the development of cutting-edge AI technologies that address real world problems confronting the video market. The platform delivers the competitive advantage that maximizes profitability for players in the video industry.
The Jump HAMLET technology was designed by Jump for the creation and maintenance of artificial intelligence models specific to the video industry. The primary objectives underpinning this new machine learning framework are (1) end-to-end control of the data flow during the generation of machine learning models, (2) the automation of processes related to the development of those models, and (3) prediction reproducibility.
In this blog post, Carlos De Las Heras Peña, Senior Data Scientist at JUMP, shares the key technical features of this machine learning -based development framework for predictive models.
The development of this ground-breaking technology is based on entry points and nodes of code, dividing the code and categorizing it according to the function it serves, rather than dividing it into individual files. This significantly increases the versatility of the code, and by combining it with computer-assisted design of dependencies, we provide development teams a tool that facilitates a declarative approach to assembling and running complete workflows, contributing to the achievement of the desired results without having to be concerned with the previous steps required.
Hamlet’s advanced code versioning is another key feature, which, thanks to its Git-based architecture, enables running each entry point using different versions from the repository. This feature simplifies the task: it is as easy as choosing to run certain entry points blocking the commit of the code version in order to ensure that your algorithms work the same way they did when they were validated in the testing phase. Another huge advantage for big development teams is the management of dependencies between entry points and nodes, since being able to select the version where you run each one ensures you are able to re-use code generated by different departments without worrying about whether they have introduced code changes. All this greatly simplifies the generation of different workflows, built by different teams to their own timescales.
Furthermore, the use of MLFlow within HAMLET enables online access to a dashboard of generated results, providing an online interface used for requesting and reviewing any defined execution metric.
Another module included in the framework is the advanced management of configuration files, which allows for the use of methods to dynamically adjust the values in these files, making it truly easy to give them virtually unlimited versatility.
Hamlet also provides flexible, automated AWS EMR support for security settings for any company, using improved configuration functionality that can launch, resize, and terminate EMR spot clusters to get the task at hand done. In addition, Spark session automatic optimization takes advantage of specific hardware features of the EMR cluster, allowing the optimization of the flows, provided Spark is being used.
In short, HAMLET provides a project creation schema, which documents the relationship between the nodes of code and entry points, making it much easier for any member of the team to understand the complete workflow, thus boosting employee performance during those crucial, intense first weeks.
With this development, Jump has met its goal of democratizing access to prohibitively expensive AI technologies and optimizing process efficiency, which otherwise would require significant amounts of economic, technical, and human resource.
In the near future, Jump will release its development framework for AI models so the entire community can contribute to and benefit from its capacity to efficiently implement AI models.