CASE STUDY · DATA SCIENCE PLATFORM
ML Arena redesign
Research uncovered why users got lost in a real-time AI competition platform — a two-week, end-to-end redesign turned the struggling MVP into an adopted product.
+1000%
registered users after the redesign
+700%
platform traffic after the homepage launch
2×
active competition participants
Role
UX Researcher · UX/UI Designer
Platform Desktop
Timeline 2 weeks · December 2024
TL;DR — THE 30-SECOND VERSION
Problem — an MVP with low adoption; the client suspected the UI, but didn't know why users weren't coming back.
Discovery — the flow logic was sound; missing orientation (no homepage, no instructions, no exits) made users feel lost.
Solution — new homepage, in-flow instructions, persistent nav and exit paths, clearer UX writing, global user rankings, full UI overhaul.
Result — +1000% registered users, +700% traffic, 2× active participants.
Overview
ML Arena is a web platform where students and data scientists compete by deploying AI agents in real-time machine learning competitions. Adoption was low and usability issues blocked engagement — I redesigned the core flow, from joining a competition to deploying a bot, plus a full UI overhaul. The work spanned UX research, information architecture, UX/UI design, and stakeholder collaboration.
Research
Seven methods, sequenced from stakeholder context to user evidence — so design decisions rested on causes, not assumptions.
Stakeholder interview. Goals, pain points, and what "low engagement" actually meant.
Competitor benchmarking. Kaggle and AIcrowd set the bar: onboarding, polish, rankings.
Heuristic evaluation. Nielsen's 10: system status, exits, recognition, and feedback all failing.
Walkthrough + deploy test. A first-user lens surfaced friction before any interviews.
User flow mapping. Proved the flow logic was sound — orientation cues were missing.
User interviews. 4 data scientists and students — all lost without an entry point.
Usability testing. Guided + unguided MVP tasks confirmed every hypothesis.
Synthesis. Affinity diagram + problem tree to cluster; MoSCoW to prioritize.
Findings
No entry point. Users landed mid-flow with zero context; everyone expected a homepage.
Lost in the flow. No persistent nav or anchors; the leaderboard was unfindable.
Zero instructions. Users guessed every step; several explicitly asked for a how-to.
Trapped after deploy. Deployment dumped users in "My Agents" with no exit but the logo.
Misleading button. "Join agent" silently created a new bot on every click.
No global rankings. Bots ranked per competition; users wanted Kaggle-style platform-wide standing.
"It's not clear what this is or where to begin."
— Ignacio, data scientist, first session on the MVP
Solutions
No homepage → New homepage. Purpose, how it works, and a clear first action.
Confusing flow → Structural anchors. Vertical nav, table of contents, redesigned game nav bar.
Zero instructions → Step-by-step guidance. "Instructions before starting" inside the Overview page.
Navigation loop → Exit paths. "Back to Upload / Competition" buttons + active-section highlighting.
Misleading button → Clear UX writing. "Create an Agent" + clarified modal copy.
No rankings → Global User Rankings. Player-based, cross-competition — #2 most-visited page after launch.
Design process
Wireframes. Lo-fi sketches → mid-fi wireframes, refined through user feedback.
Brand attributes. Efficient · Modern · Elegant — proposed and approved with the client.
Palette & type. Navy + gray, purple/bright-blue accents; Inter body, Orbitron logo.
Key learnings
Research first.
It anchored design in real causes — not the client's assumption that UI was the problem.
Simple fixes win.
No flow rebuild needed; targeted UX elements unlocked adoption.
Entry points earn trust.
The missing homepage was the biggest break with user expectations.
The stakeholder credited the redesign as a key driver of product adoption and overall improvement.
Let's work together
CASE STUDY · DATA SCIENCE PLATFORM
ML Arena redesign
Research uncovered why users got lost in a real-time AI competition platform — a two-week, end-to-end redesign turned the struggling MVP into an adopted product.
+1000%
registered users after the redesign
+700%
platform traffic after the homepage launch
2×
active competition participants
Role UX Researcher · UX/UI Designer
Platform Desktop
Timeline 2 weeks · December 2024
TL;DR — THE 30-SECOND VERSION
Problem — an MVP with low adoption; the client suspected the UI, but didn't know why users weren't coming back.
Discovery — the flow logic was sound; missing orientation (no homepage, no instructions, no exits) made users feel lost.
Solution — new homepage, in-flow instructions, persistent nav and exit paths, clearer UX writing, global user rankings, full UI overhaul.
Result — +1000% registered users, +700% traffic, 2× active participants.
Overview
ML Arena is a web platform where students and data scientists compete by deploying AI agents in real-time machine learning competitions. Adoption was low and usability issues blocked engagement — I redesigned the core flow, from joining a competition to deploying a bot, plus a full UI overhaul. The work spanned UX research, information architecture, UX/UI design, and stakeholder collaboration.
Research
Seven methods, sequenced from stakeholder context to user evidence — so design decisions rested on causes, not assumptions.
Stakeholder interview. Goals, pain points, and what "low engagement" actually meant.
Competitor benchmarking. Kaggle and AIcrowd set the bar: onboarding, polish, rankings.
Heuristic evaluation. Nielsen's 10: system status, exits, recognition, and feedback all failing.
Walkthrough + deploy test. A first-user lens surfaced friction before any interviews.
User flow mapping. Proved the flow logic was sound — orientation cues were missing.
User interviews. 4 data scientists and students — all lost without an entry point.
Usability testing. Guided + unguided MVP tasks confirmed every hypothesis.
Synthesis. Affinity diagram + problem tree to cluster; MoSCoW to prioritize.
Stakeholder interview. Goals, pain points, and what "low engagement" actually meant.
Competitor benchmarking. Kaggle and AIcrowd set the bar: onboarding, polish, rankings.
Heuristic evaluation. Nielsen's 10: system status, exits, recognition, and feedback all failing.
Walkthrough + deploy test. A first-user lens surfaced friction before any interviews.
User flow mapping. Proved the flow logic was sound — orientation cues were missing.
User interviews. 4 data scientists and students — all lost without an entry point.
Usability testing. Guided + unguided MVP tasks confirmed every hypothesis.
Synthesis. Affinity diagram + problem tree to cluster; MoSCoW to prioritize.
Competitor benchmarking. Crimpd and
Social Boulder — both static, single-user,
gym-bound experiences.
Survey — 38 climbers. 65% tracked progress
informally; 81% wanted structure; 74%
motivated socially.
7 user interviews. Intermediate climbers in
France — already tracking, but in scattered,
improvised ways.
Persona & user flows. Lucas, the community-
motivated tracker — grounding every flow
decision.
Usability testing. Wireframes and prototype
tested and iterated with climbers through the
sprint.
Synthesis. Affinity diagram to cluster pain
points; MoSCoW to scope the two-week MVP.
Findings
No entry point. Users landed mid-flow with zero context; everyone expected a homepage.
Lost in the flow. No persistent nav or anchors; the leaderboard was unfindable.
Zero instructions. Users guessed every step; several explicitly asked for a how-to.
Trapped after deploy. Deployment dumped users in "My Agents" with no exit but the logo.
Misleading button. "Join agent" silently created a new bot on every click.
No global rankings. Bots ranked per competition; users wanted Kaggle-style platform-wide standing.
No entry point. Users landed mid-flow with zero context; everyone expected a homepage.
Lost in the flow. No persistent nav or anchors; the leaderboard was unfindable.
Zero instructions. Users guessed every step; several explicitly asked for a how-to.
Trapped after deploy. Deployment dumped users in "My Agents" with no exit but the logo.
Misleading button. "Join agent" silently created a new bot on every click.
No global rankings. Bots ranked per competition; users wanted Kaggle-style platform-wide standing.
"It's not clear what this is or where to begin."
— Ignacio, data scientist, first session on the MVP
Solutions
No homepage → New homepage. Purpose, how it works, and a clear first action.
Confusing flow → Structural anchors. Vertical nav, table of contents, redesigned game nav bar.
Zero instructions → Step-by-step guidance. "Instructions before starting" inside the Overview page.
Navigation loop → Exit paths. "Back to Upload / Competition" buttons + active-section highlighting.
Misleading button → Clear UX writing. "Create an Agent" + clarified modal copy.
No rankings → Global User Rankings. Player-based, cross-competition — #2 most-visited page after launch.
Design process
Wireframes. Lo-fi sketches → mid-fi wireframes, refined through user feedback.
Brand attributes. Efficient · Modern · Elegant — proposed and approved with the client.
Palette & type. Navy + gray, purple/bright- blue accents; Inter body, Orbitron logo.
