Predictive Analytics is a machine learning service on Bluemix geared towards developers. It allows for live processing of data streams through a predictive model via API requests, for such uses as fraud detection, making production recommendations or predicting customer churn.

Getting from Beta to General Availability

The Predictive Analytics service had launched on IBM's Bluemix in a beta release. As a condition for a full, general availability release, any service on Bluemix has to conform to a set of design guidelines in order to provide for good user experience. Unfortunately, due to miscommunication, the service didn't match the design requirements, and didn't have any designers to work on the product; putting the service's upcoming general availability release into peril.

I came to the product with just a few weeks before launch. I took the existing design andupdated some of the UX patterns to better fit a user's expectation of functionality. I also worked with a visual designer on my team, to incorporate design guidelines from IBM's design language as well as my group's own internal style guide. I worked closely with offering management and development to make sure design was correctly executed, as well as with Bluemix, so the service could get launched in time.

Before: Exisiting beta design

After: Updated design for general availability release

Creating the next version

Mapping out the golden path / hero flow for building a machine learning model

The first release of Predictive Analytics allowed developers to upload a predictive model, created in SPSS by a data scientist, to the cloud so they could use it for real time predictions via API.

But requiring a data scientist to create the model was a pain point, and competition like AWS and Google Prediction API allowed developers to build machine learning models on their own.


Goal: Allow a developer to build a machine learning model within their browser, without the need for an experienced data scientist.

As a designer, I conducted competitive research and mapped out a user scenario to allow a developer to build their model.

I also thought how future capabilities of batch processing of models and version management would be incorporated into the UI.


Wireframes: All Screens


Key Flow: New Auto Machine Learning Model

I constructed step by step wire frames of a flow for what building a new, machine learning model would like in the context of IBM Bluemix.