Prediction Service
Automated model creation for effortless and optimal ML model generation
The Product
Automated ML from Data to Deploy
Prediction Service enables training of custom ML models with minimal effort and machine learning expertise. It provides a simple interface to create, train, and deploy models.
My Role
Refactoring and expanding
I led the redesign of Prediction Service taking it from beta version to first launch. I refactored the information architecture and navigation to optimize workflows and add new features.
Outcomes & Impact
Growth and Adoption
The redesign of Prediction Service enabled users to create and deploy models faster, with fewer errors, resulting in an increase in user engagement and an increase in NPS scores.
Context
Initial launch showed success for training models, however users struggled with managing and deploying models
AutoML streamlines the most tedious tasks of model training, so with the launch of Prediction Service users were able to train models with minimal effort.
However, users still struggled to deploy and manage models, leading to the impetus for the redesign and expansion of Prediction Service to include end-to-end workflow of model training, and deployment.
Screen designs
Below are a few screens from Prediction Service redesign
New features included the deployment of models, viewing and comparing experiment results, scheduling and managing automated and ad-hoc jobs, viewing model details, model history, and guardrail configuration.
