Model-Training Hub

Model Training Hub

Designing an AI model training platform to transform fragmented engineering workflows into structured, interpretable processes.
Focused on making model training legible, navigable, and cognitively manageable for both advanced and basic users.
Reduced task completion time by 68% and increased user satisfaction by 139%.

Scope

Redesign of AI model training platform
Workflow and task-flow analysis
Persona definition and usability testing
Rapid prototyping and interaction modeling
Visual system refinement
Documentation and engineering handover

My Role

Led the end-to-end redesign of the training experience
Translated machine learning concepts into structured product logic
Defined UX roadmap and design direction
Designed interaction models and prototypes
Measured usability impact through testing and iteration
Collaborated closely with ML PMs, engineers, and data teams

About Project

The platform enables teams to train and manage AI models.
When I joined, it was technically powerful but structurally difficult to navigate. Engineers had prioritized MVP delivery under time pressure, resulting in fragmented workflows and limited visibility into training progress.

The redesign focused on making the training process interpretable, not just executable.

The Challenge

The system functioned, but it lacked:

Clear process visibility
Coherent training flow
Role-specific mental models
Transparent system status.

Users struggled not because the system lacked capability, but because it lacked structural clarity.

The objective was to redesign the platform so users could understand where they were in the training lifecycle, what decisions were required, and how model progress evolved over time.

Users and Pain points

Advanced Users

Inefficient navigation across fragmented pages
Low visibility of system state and training progress
Increased risk of errors due to hidden status signals

Basic Users

An incomplete mental model of the training process
Uncertainty about next actions
Difficulty understanding competency thresholds and required inputs

Key Questions

How can the training lifecycle be made structurally visible?
How can users understand model progress without technical abstraction?
How can system status be surfaced without overwhelming users?

Process

I worked in rapid iteration cycles: paper sketches, digital prototypes, usability testing, and direct feedback from AI engineers.

Each iteration revealed deeper insights into how users interpret training states, transitions, and model feedback. The goal was not incremental UI improvement; it was restructuring how the system communicates progress, uncertainty, and decision points.

Design Approach

focused on designing a coherent training journey rather than isolated screens.

Instead of adding scattered tooltips or navigation elements, I restructured the platform around:

A clearly staged training process
Persistent visibility of system state
Reduced context-switching
Defined decision checkpoints

Training was reframed as a guided, transparent progression rather than a set of disconnected technical actions.

DESIGN DECISIONS

1. Overview Architecture

Introduced a centralized overview page summarizing:

Model status
Training stages
Key metrics
Pending actions

This reduced navigation friction and provided users with a single cognitive anchor for understanding system state. The overview functioned as a structural dashboard, not decorative, but epistemic.

2. Dedicated Training Workspace

Designed a focused training guide page with a persistent work area. This eliminated excessive page switching and supported continuous, iterative model refinement. The interface reinforced progression, reducing ambiguity in what to do next and where users stood in the lifecycle.