Loading
Skip to main content
Case Study 002

Democratizing Machine Learning for Everyone

Turning a black-box data science process into a guided, 4-step workflow.

// STATUS: SHIPPING_2026
AI/ML_UXSystems_ThinkingZero-to-OneCross-functional_LeadershipTechnical_Fluency
ml_wizard_pipeline.py
ML Functions Confusion Matrix - Model Performance Visualization
UX Lead / UX Owner (End-to-End Ownership)·Cloud Software Group — WebFOCUS·Jan 2023 – Jan 2025 | Shipping 2026

The best screen in the entire UX revamp.

— Principal Data Scientist, on the confusion matrix I designed after 10+ iterations together

SME Validation

At a Glance

Situation

A fragmented ML training workflow with 12+ clicks, hidden hyperparameters, and dead-end error states. It worked for data scientists but wasn't accessible to business users.

Task

Make machine learning accessible to non-technical users without dumbing it down for experts.

Action

Got MIT certified in AI/ML product design. Embedded weekly with the Principal Data Scientist for months. Navigated 2-year timeline through engineering delays and layoffs. Hands-on with engineering due to limited front-end resources.

Result

Reduced workflow from 12+ clicks to 7-9. All five SMEs found the entry point without help. Our Principal Data Scientist called the confusion matrix "the best screen in the entire UX revamp."

Sound On · Director's Commentary

The Transformation

Narrated walkthrough of the Legacy friction vs. the Modern unified workflow.

Legacy ML Workflow

The old fragmented workflow: 4+ step path, drag model pill onto data flow, configure in popup, hidden hyperparameters, confusing "results not generated" errors.

New Guided Workflow

The new guided workflow: structured step-by-step flow, right-click entry, clear error handling, reduced from 12+ clicks to 7-9 clicks.

Password Required

Click to unlock redesigned workflow

Key Differences

Before
  • Fragmented 4+ step workflow
  • 12+ clicks through data flows and menus
  • Hidden hyperparameters behind right-click
  • Confusing "results not generated" errors
After
  • Structured 4-step guided flow
  • 7-9 clicks via right-click entry
  • Clear error handling and validation
  • Accessible for non-technical users

Discover Deeply: How I Earned the Project

TL;DR

Turned a knowledge gap into an advantage. MIT certified. Months learning before designing.

Zero ML background could have been a liability — instead, it became an advantage. I enrolled in MIT's AI/ML product design certification and embedded weekly with our Principal Data Scientist.
By the time I started designing, I understood the domain deeply enough to challenge assumptions and ask the right questions.

Empathize with the Ecosystem: Understanding Users and Workflows

TL;DR

Three personas with conflicting needs. The existing experience served none of them well.

Three personas. Three different needs: data scientists wanted depth and control over hyperparameters, business users wanted simplicity and guidance, analysts wanted both. The existing experience served none of them well.
After documenting every workflow and decision point, I realized: if I find this frustrating after weeks of study, a first-time user has no chance. That insight drove the redesign.

Simplify the Chaos: From Black Box to Guided Flow

TL;DR

Designed a structured 4-step guided flow: Problem Type → Target → Predictors → Hyperparameters.

The mapping revealed the problems, but solving them required multiple iterations. Three critical pivots shaped the final design.
Breakthrough: I asked our Principal Data Scientist, "What do you absolutely need to train a model responsibly?" His answer: problem type, target variable, predictors, and hyperparameters. That became the 4-step UX spine — a linear flow that made ML accessible without dumbing it down.
// ARCHIVE_EVIDENCE

The System Blueprint

Before solving for UI, I had to solve for system logic. These diagrams map the complex relationships between model types, training workflows, and user personas.

Full Access

View Original PDF

Early concept wireframes

Early concept wireframes

System Notes

System Notes

User Flow Mapping

User Flow Mapping

Architecture Diagram

Architecture Diagram

Logic Map

Logic Map

Early Wireframes

Early Wireframes

+5

more pages in the full archive

🔒

Design Pivots

Detailed artifacts and sensitive diagrams are available for authorized reviewers.

// UX_PRINCIPLES

Design Principles Applied

Explainability FirstLinear Flow (No Branching)Upstream ValidationSystem State VisibilityProgressive DisclosureDual Experience (Guided + Advanced)
🔒

Design Iterations

Detailed artifacts and sensitive diagrams are available for authorized reviewers.

🔒

Layered Disclosure Strategy

Detailed artifacts and sensitive diagrams are available for authorized reviewers.

🔒

Design Artifacts

Detailed artifacts and sensitive diagrams are available for authorized reviewers.

Grow Through Constraints: Earning Trust & Leading the Team

TL;DR

Dual-experience approach emerged from constraints. Led this while simultaneously owning ReportCaster and IQ Plugin.

The dual-experience approach emerged from engineering constraints. We couldn't rebuild the advanced mode — it had to coexist with the new guided flow. What felt like a limitation became a feature: experts got their power, newcomers got guidance.
I led this while simultaneously owning ReportCaster and IQ Plugin. Cross-project pattern sharing meant solutions in one project accelerated the others.

Navigate Forward: Impact, Validation, and Reflection

TL;DR

5/5 SMEs found entry without help. Patterns became foundation for IQ Plugin and platform-wide AI strategy.

Key Metrics:5/5
5/5 SMEs found the entry point without help. Dead-ends → clear guidance. Design demos to 150-200 person business unit earned leadership support.
The patterns I developed here — structured flows, upstream validation, right-click entry, dual-experience — became the foundation for IQ Plugin and platform-wide AI strategy.
// VISUAL_DIFF

Legacy vs. Redesigned Workflow

🔒

Protected Content

Detailed artifacts and sensitive diagrams are available for authorized reviewers.

Interested in working together?