Project Overview
Argon's co-founders built a promising AI research assistant for healthcare and biotech analysts, but only they could navigate it. With no design input, it felt more like a troublesome intern than a helpful assistant. As the sole designer, I rebuilt the experience so researchers could trust it and use it without a guided tour.
Role: Lead & sole UX designer
Team: 2 co-founders, PM, developer
Timeline: ~3 months
Outcome: Core patterns I designed, including structured AI queries and AI-generated columns, are still central to the live product.
What Was Broken
I started with a heuristic audit of the MVP. Users couldn't tell where to type, whether the pills were filters or tags, or what the AI would do with their question. Nothing gave feedback, and nothing had hierarchy.
Final Designs - Restructuring the entry point
New users didn’t know what to ask or how to phrase it. I replaced the free-text prompt with a category dropdown, which gives people a clear starting point and gives the AI more precise input. That was a win for the founders too.
New Feature: Selecting Data Sets
For high-stakes research, users need to know where answers come from. After the AI’s quick answer, users can “dive deeper” by picking specific sources (analyst reports, news, clinical studies), with a short abstract for each so they don’t have to remember what’s in every data set.
From a static table to a working space
In the MVP, a “project” was just a static table. I turned it into a workspace with three resizable panels (source data, AI insights, and visual outputs) so researchers can arrange it around what they’re doing. An “Add Column” panel lets them ask new questions mid-project and preview the AI’s answers before adding them to the table.
Design System - Material Design 3
A system built to flex
I built a modular design system (type, color, icons, components) tuned for a dense, data-heavy interface. Mid-project, engineering chose Material UI, so I re-mapped everything to Roboto Flex and MUI’s color tokens. Because the system was already modular, updating the entire platform took about a day.
Reflections
Good AI UX isn’t about making the model look impressive. It’s about helping people ask better questions and trust what comes back. With more time, I’d test the structured query flow and workspace with first-time researchers to see where they still get stuck.
Check out my next case study