Platform-wide orchestration

Unlike any tool of its kind, this canvas based UI will enable Data Scientists and Machine Learning Operations Engineers to automate operations between platform assets to control end to end lifecycle processes. The tool will be built to be compatible with the open-source Kubeflow tool for easy user adoption. The tool is currently in development and has not been released to the public.

I was the lead designer for the orchestration editor tool that will appear in Watson Studio and Cloud Pak for Data at the end of 2020. I directed weekly syncs to showcase the evolution of the design to a cross-functional team of Developers, System Architects, Data Scientists, and Product Managers across four time zones (from San Francisco to Poland).

I created and facilitated a design session for our cross-functional team via Mural to begin aligning on a vision. The tightly packed agenda of six exercises in one hour allowed a forum for 20 teammates to ultimately reveal their vision without technical constraints. All the exercises were chosen based on the nature of the event: virtual, fast, and direct. Below is an overview of some of the exercises:

1. Hopes and Fears: Team members start the dialog by opening up about insecurities which directly squashes imposter syndrome by revealing other teammates have similar concerns. This built trust amongst a new team and opened a dialogue on the product vision.

2. Assumptions and Questions: Because our team was spread across four time zones it was important to note what was known and not known about the project. Communication continued to be the biggest challenge throughout the project and by starting here, we all began on the same playing field.

3. As-is journey mapping: Ending the first half of the design session, we mapped how users accomplish their goals currently, with no solution in place. This revealed pain points that I addressed when it was time for me to prototype.

At this stage, I learned basic answers to these questions that allowed me enough understanding to start prototyping.

Who: Data Scientists and ML Ops Engineers

What: Combine a number of data analytics assets and actions (such as data, databases, data flows, ML models, accuracy checks, etc) in order to build a flow across many of IBM’s existing tooling

When: After a user has created analytic assets that they want to use in active learning, learning, model deployment, or model training

Where: Accessed from either Cloud Pak for Data or Watson Studio

Within IBM Data and AI, there are four canvas based tool teams ranging in maturity from two to 10 years old. Each team is based on a common canvas which gets customized per the tools individual needs. I reached out to these teams for feedback which allowed me to understand the design system through various applications of a canvas tool in different scenarios.

I began wireframing by combining our priorities from the design session and my knowledge from the canvas teams and sketched several user flows with different considerations such as:

After receiving feedback from both my team and the canvas teams across IBM Data and AI, I updated and designed an MVP flow for my team to review via Invision. Because my team had challenges meeting due to extreme differences in time zones, the Invision prototype allowed feedback and discussions to happen at any time, which sped up the process.

This project was handed off to another designer to execute user testing, final iteration, and hand off to developers to be released at the end of 2020.

This project will result in: