Centurylink Model Controller

Redesigned a data platform for managing predictive models, improving visibility and efficiency for data scientists

Tools: Adobe Creative Suite, Figma

The Problem

Data scientists were building predictive models in Python and R, but lacked a centralized way to manage, track, and evaluate them.

Each model required:

  • manual execution

  • individual tracking

  • limited visibility across teams

This created inefficiencies and made it difficult to compare performance across models or deployments.

Why this was challenging

  • Highly technical user base (data scientists)

  • Complex datasets and model outputs

  • Need to support multiple model deployments and versions

  • Product direction shifting frequently due to consulting-based origins

  • Lack of consistency across features and workflows

Approach

We focused on transforming the experience into a centralized, dashboard-driven platform.

Key priorities:

  • Provide a single source of truth for model management

  • Enable comparison across models and deployments

  • Surface meaningful metrics through clear visualizations

  • Support organization of models into projects and workflows

The goal was to simplify complex technical processes without limiting flexibility for advanced users.

Solution

We designed a desktop application that allowed data scientists to:

  • Upload and manage predictive models in Python and R

  • Track multiple deployments of the same model

  • View detailed success and failure metrics through D3-powered visualizations

  • Organize models into projects for easier management

  • Collaborate across teams with role-based access and permissions

The interface focused on clarity and structure, making complex data easier to interpret and act on.

Log In Screen

  1. Using Centurylink log in credentials, a user can log in directly to the portal beginning with their team name.

Home Dashboard

  1. A user is able to select which server to view.

  2. A comprehensive numerical summary of all models that are completed, failed or currently running.

  3. A list of all projects on any given server as well as a list of any model executions running in the last 24 hours.

Projects

  1. From the “Projects” page a user can see a list of all projects assigned to them. Each project can contain a variety of models/deployments.

Add Project Modal

  1. A user may add a new project and enter a project name, description and upload any models/deployments.

Edit Existing Project

  1. At any time a user can open an existing project and edit any of the data, most importantly removing or adding any new deployments.

Model Dashboard

  1. Upon page load the user can see a graph of the model deployments with submenus for specific version uploads, scheduling, any uploaded documentation and specific settings/parameters for the executions.

  2. A list of recent executions for that particular model/deployment, timestamped and indicating whether the model is completed, running or failed.

Product Evolution

Originally developed as a consulting tool, the product evolved through multiple client engagements.

This created challenges:

  • Frequent feature changes

  • Shifting priorities

  • Lack of a cohesive core experience

Through iterative design, we worked to unify the product and establish consistent patterns across workflows.

Status

This project was developed as an internal platform initiative and validated through user testing but was ultimately not brought to full release due to shifting business priorities.

Outcome

The redesigned platform demonstrated a strong foundation for centralized model management:

  • Simplified how data scientists could upload and manage predictive models

  • Improved visibility into model performance through structured dashboards

  • Enabled better organization and collaboration across teams

User testing within Centurylink showed the interface was well-received and aligned closely with existing internal tools, allowing it to integrate naturally into the broader ecosystem.

While the product was not brought to full release, the work established a scalable framework for managing predictive models and informed future internal platform efforts.