Predict Data – Making Machine Learning Accessible

I began by immersing myself in ML fundamentals—partnering with our data scientist on weekly deep‑dive sessions, collaborating closely with the lead engineer, and completing an MIT AI & ML course. Armed with that knowledge, I redesigned the fragmented, portal‑hopping workflow into a seamless, hub‑based experience: right‑click any dataset, launch Predict Data, and follow a clear, step‑by‑step guide. Early usability tests showed a 40% reduction in task time and far fewer errors, proving that combining UX clarity with functional depth truly empowers users.

Impact at a Glance

100% Discoverability

All testers found “Predict Data” on first try

2-Click Entry

Right-click → Predict Data vs. 4+ clicks before

50% Faster Setup

Model build time halved in usability tests

60% Fewer Errors

After two built-in tutorials

The Problem

Fragmented & Hidden: ML tools were buried in a separate portal UI, forcing context switches and making discovery nearly impossible.
No Clear Entry Point: Users faced 4+ unintuitive steps just to begin training—navigation, drag‑drops, and switches left them guessing where to start.
Confusion & Frustration: Screens lacked labels or guidance, so users didn’t know what each step meant or how to interpret results, leading to frequent support tickets.
Opaque Workflow: The interface offered no explainability or step‑by‑step guidance, intimidating non‑technical users and undermining confidence in ML.

Competitive Landscape

Understanding the Users

Designed for data analysts and business stakeholders—balancing advanced controls with plain-English guidance.

Goals

Bring ML to the user’s fingertips — Instead of forcing analysts into a separate portal, let them right-click any dataset and launch Predict Data instantly—no extra tabs, no hidden menus.
Guide every step of the journey — Replace a maze of disconnected screens with a coherent, step-based wizard that holds your hand from “select data” through “interpret results.”
Empower non-technical champions — Build in just-in-time tutorials, plain-English labels, and contextual help so business users can train their first model with confidence—no PhD required.
Balance power and simplicity — Surface advanced controls for data scientists alongside friendly defaults for casual users, so everyone finds the right level of detail at the right time.

Design Evolution

Entire ML workflow

ML workflow user-based

Step-by-step workflow entry

High-fidelity prototype

My Role

Owned End-to-End UX – Volunteered as the sole designer to transform an undocumented, engineer-driven ML interface into a user-centered workflow.
Up‑skilled on the Fly – Completed MIT & Amazon AI courses and ran weekly deep‑dive calls with data scientists to translate ML concepts into design.
Iterated Relentlessly – Crafted 25+ workflow versions, each refining clarity—balancing power‑user controls with plain‑English guidance.
Aligned Cross‑Functionally – Partnered with engineers, QA, PMs, and SMEs to validate feasibility, catch edge cases, and champion the UX internally.
Ensured Pixel‑Perfect Delivery – Conducted implementation audits to guarantee shipped HTML/CSS matched hi‑fi designs and reduced UI inconsistencies.
Championed Educative UX – Hosted SME user‑groups, embedded contextual help, and turned black‑box ML training into a transparent, learning‑first experience.

Internal Feedback & Next Steps

SMEs and customer reps praised its ease; public beta Q1 2026, with advanced analytics instrumentation to follow.

Workflow Overview

Landing page (Step 1)

Workflow steps 2 & 3

Cine Compare Model UI

Train Model Results

Optimize Model & Run Models Review