MuchSkills

AI Workspace

The AI Workspace is a tool that facilitates the creation and deployment of artificial intelligence and machine learning workflows.

It is designed to be a drag-and-drop interface, in an attempt to simplify the underlying tech for the end user.

Between Sep '19 - Aug '20, I had the opportunity to collaborate with the AthenasOwl↗ team as a product designer.

During this time, I led the design on an AI enabled content creation tool↗ as well as on a comprehensive design system↗

Below is a high level compilation of all the work I was able to do on the AI Workspace, which is a tool that facilitates the creation and deployment of artificial intelligence and machine learning workflows. It is designed for teams that work with sports content to parse videos and add information to them to make them more valuabe.

To honor non-disclosures, I must withhold any and all information about the design process -- but feel free to reach out to me and I would be happy to walk you through a deeper dive.

Roles Design Strategy, Service Design, UX/UI Design

Duration Sep '19 - Aug '20

Tools Sketch, Principle

Video Pre-Processing

A machine reads a video as a sequence of frames (images) strung together. For a machine to read a video and extract information from it, each video has to be processed.

Videos are first broken down into smaller videos (clips), and then frames (images) using standard machine learning processes. The resultant frames can then be read in many ways depending on the kind of information that needs to be extracted.

Looking for a person/player? Face recognition.

Looking for an object? Entity detection.

Therefore, the processes that need to run on a video vary depending on what the output has to be. This creates a need for variable video processing workflows, and the AI Workspace facilitates the creation of these variable workflows.

The Canvas The interface is designed as a canvas of sorts, to simplify the underlying tech for the end user (someone most likely to not be well versed with AI/ML).

Workflow Creation Workflows are easy to create, by simply dragging and dropping processes onto the canvas from the toolbox on the left.

Workflow Scheduling Workflows are generally triggered at the occurence of an event (video available) or at pre-determined times (post broadcast). The AI Workspace enables such event or time based scheduling.

Workflow Trigger & Track Once a workflow is triggered, each process is executed linearly or in parallel, and a tracker allows the user to monitor progress.

Scheduling on the go The mobile version of the tool makes it easy for users to schedule pre-made workflows from wherever they are.

Tracking on the go A lot of the processes take some time to be executed completely. Users can use the mobile version of the tool to keep an eye on the progress at any given time.

What next?

While the video pre-processing aspect of the AI Workspace takes care of all the manual work that content teams have to do (scanning and tagging videos), I wanted to explore ways in which we could take this experience to the next level.

I worked on a concept which allowed users to visualise the extracted data from each processed video, and proceed to edit it the way they would like.

I did not want the users of this tool to be reduced to mere screen watchers / workflow trackers, and felt like this was an interesting way to introduce another layer of human interaction within the tool.

We were able to execute some very successful video pre-processing projects using the AI Workspace, particularly with clients who dealt with broadcasting football matches.

In the time I spent as a part of the team, we would use the tool to generate metadata and share that with each respective client. A very interesting use case we were able to solve was detecting the amount of time a particular player was seen at the same time as a particular logo.

Unfortunately, I stepped away from the team before we could dive into the video editing concept, which as per my understanding has not yet been realised.