Johanna Huarachi's Portfolio*

TLDR

Overview

I redesigned the filtering system for Expert.ai Corpus, an AI text analysis platform. My work focused on replacing a confusing popup with a side panel that lets users filter while viewing results—reducing support tickets by 42% and cutting task time from 2 minutes to 30 seconds.

ROLE

UX Design Intern

ROLE

UX Design Intern

ROLE

UX Design Intern

TIMELINE

6/1/2022 - 8/31/2022

TIMELINE

6/1/2022 - 8/31/2022

TIMELINE

6/1/2022 - 8/31/2022

TEAM

Johanna Huarachi

Antonio Linari

Luigi G

TEAM

Johanna Huarachi

Antonio Linari

Luigi G

TEAM

Johanna Huarachi

Antonio Linari

Luigi G

SKILLS

User research

Interaction design

Accessibility

SKILLS

User research

Interaction design

Accessibility

SKILLS

User research

Interaction design

Accessibility

PROBELM

AI can analyze thousands of documents instantly. But users couldn't use our filters.

Expert.ai uses AI to analyze massive text data—work that would take humans weeks. But our filtering system wasn't keeping up. 62 support tickets in 6 months. Users couldn't tell which filters were active, drag-and-drop felt clunky, and the popup blocked their results entirely.

AI can analyze thousands of documents instantly. But users couldn't use our filters.

How can we make filtering as intuitive as the AI behind it?

RESEARCH

Current filtering and what users told us

The filter popup blocked users' results entirely. They'd add a filter, lose sight of their data, close the popup to check, reopen it to adjust... repeat 4-5 times per session.

I reviewed 127 support tickets and talked to 10 enterprise users across healthcare, finance, and government. Here's what they said:

Key Insight

Users needed to see their data while filtering it. Our interface forced them to choose between the two.

Key Insight

Users needed to see their data while filtering it. Our interface forced them to choose between the two.

Key Insight

Users needed to see their data while filtering it. Our interface forced them to choose between the two.

IDEATION

Redesign the flight-search experience with persistent navigation, geographic context, and scannable information.

From safe fixes to breaking the design system
We started conservative—better labels, improved colors. I created two variations within our existing design system. Both looked great. Both failed testing because users still couldn't see results while filtering. We had to rethink the entire interaction.
The Side Panel
What if filters sat next to results instead of covering them? Users could see data update in real-time, no more popup toggling, and simple on/off switches instead of drag-and-drop. The challenge: This broke our design system. Longer timeline, more engineering work. I had to prove it was worth it through testing.
The Dropdown
We also explored a dropdown menu approach—more feasible, closer to our system. But users wanted to see all filters at once, not hunt through menus. The side panel won because it solved visibility while making filtering faster.
From safe fixes to breaking the design system
We started conservative—better labels, improved colors. I created two variations within our existing design system. Both looked great. Both failed testing because users still couldn't see results while filtering. We had to rethink the entire interaction.
The Side Panel
What if filters sat next to results instead of covering them? Users could see data update in real-time, no more popup toggling, and simple on/off switches instead of drag-and-drop. The challenge: This broke our design system. Longer timeline, more engineering work. I had to prove it was worth it through testing.
The Dropdown
We also explored a dropdown menu approach—more feasible, closer to our system. But users wanted to see all filters at once, not hunt through menus. The side panel won because it solved visibility while making filtering faster.
From safe fixes to breaking the design system
We started conservative—better labels, improved colors. I created two variations within our existing design system. Both looked great. Both failed testing because users still couldn't see results while filtering. We had to rethink the entire interaction.
The Side Panel
What if filters sat next to results instead of covering them? Users could see data update in real-time, no more popup toggling, and simple on/off switches instead of drag-and-drop. The challenge: This broke our design system. Longer timeline, more engineering work. I had to prove it was worth it through testing.
The Dropdown
We also explored a dropdown menu approach—more feasible, closer to our system. But users wanted to see all filters at once, not hunt through menus. The side panel won because it solved visibility while making filtering faster.
From safe fixes to breaking the design system
We started conservative—better labels, improved colors. I created two variations within our existing design system. Both looked great. Both failed testing because users still couldn't see results while filtering. We had to rethink the entire interaction.
The Side Panel
What if filters sat next to results instead of covering them? Users could see data update in real-time, no more popup toggling, and simple on/off switches instead of drag-and-drop. The challenge: This broke our design system. Longer timeline, more engineering work. I had to prove it was worth it through testing.
The Dropdown
We also explored a dropdown menu approach—more feasible, closer to our system. But users wanted to see all filters at once, not hunt through menus. The side panel won because it solved visibility while making filtering faster.

THE SOLUTION

Side panel keeps users in flow

After testing with 8 users, the side panel was the clear winner. Task times dropped to 30-45 seconds and workflow abandonment disappeared.

Users see results update as they filter

Filters sit alongside results. Users watch the AI re-analyze in real-time and see if they're filtering too narrowly or just right.

Toggles replace drag-and-drop

Simple on/off switches. One click to activate, one to deactivate. Clear visual states with color, icons, and labels.

100% accessible

Every element meets WCAG 2.1 AA. Multiple visual cues (not just color) for filter states. Full keyboard navigation support.

LEARNINGS

Key Takeaways

"Small" features aren't small when users touch them constantly

I thought this was just a filter redesign. Turns out, filtering was the most-used interaction in the entire platform. What seemed like a minor UI issue was actually breaking the core experience. Now I ask: how often will users interact with this?

User data is your best weapon for risky changes

Breaking our design system was a tough sell. But 62 support tickets, 2-minute task times, and user testing videos made the case stronger than any design argument could. When stakeholders pushed back, I showed them recordings of users giving up mid-task.

OTHER WORKS

Overview

AI can analyze thousands of documents instantly. But users couldn't use our filters.

AI can analyze thousands of documents instantly. But users couldn't use our filters.

AI can analyze thousands of documents instantly. But users couldn't use our filters.

Current filtering and what users told us

Key Insight

Key Insight

Key Insight

Redesign the flight-search experience with persistent navigation, geographic context, and scannable information.

From safe fixes to breaking the design system

The Side Panel

The Dropdown

From safe fixes to breaking the design system

The Side Panel

The Dropdown

From safe fixes to breaking the design system

The Side Panel

The Dropdown

From safe fixes to breaking the design system

The Side Panel

The Dropdown

Side panel keeps users in flow

Users see results update as they filter

Toggles replace drag-and-drop

100% accessible

Key Takeaways

"Small" features aren't small when users touch them constantly

User data is your best weapon for risky changes

← DNC