AI-driven product discovery for e-commerce

I joined Moda AI while enrolled in my UX/UI bootcamp. It was a rare situation: real product design work alongside structured learning. I was the only designer on the team, which meant I owned everything visual, from early concepts to stakeholder-ready prototypes.

Over 8 months, I collaborated directly with engineers and leadership to design a conversational shopping experience. The work ranged from user research to interface design to brand identity, all happening at startup speed.

We ran 5 user interviews with online shoppers to understand how they search, where they get overwhelmed, and what they wish worked better. We also partnered with a major e-commerce retailer to align the product with real platform needs.

The key finding: users liked the idea of conversational product discovery, but didn’t trust it to work well yet. They’d been burned by bad chatbots. That told us something critical: we couldn’t just build a better chatbot. We needed to meet users where they already were, with interfaces that felt familiar first and intelligent second.

To make AI-powered product discovery accessible and adaptable, I explored four distinct interfaces. Each was tailored to a different user mindset and integration need.

The first interface was a floating chat assistant embedded within product listing pages. Think Intercom, but for shopping. Users could ask “I need an outfit for a wedding in Santa Barbara” without leaving the page they were on.

I designed it with Moda’s branded icon and a supportive tone. The key design decision was making it dismissible and non-blocking. The retailer’s existing product grid stayed intact. AI was an overlay, not a replacement.

The second interface placed a minimized ShopGPT tab alongside the product listing page. Subtly branded, easy to ignore. When expanded, it pushed the listings down and opened a full AI search experience.

This gave users complete control over when to engage with AI. The expandable pattern made the AI feel optional rather than forced, which directly addressed the trust issue from research.

The third approach was the most seamless. AI and traditional search shared a single input field, with a simple toggle to switch between them. No new layout, just new behavior underneath a familiar pattern.

Only when users chose ShopGPT did the experience transition into a conversational results page. This was the version the retailer responded to most, because it required the least disruption to their existing UI.

The fourth interface was the most ambitious: a fully conversational experience for users starting with a vague idea. “I need a dress for a beach wedding.” The interface responded with smart product suggestions organized by category, embedded visuals, follow-up questions, and a saved items feature.

This was the most immersive version of Moda’s vision. It showed what conversational shopping could look like when you give it room to breathe.

Moda needed to feel intelligent without being intimidating. I built a soft, modern visual system using Poppins for typography and a blue-to-purple gradient that felt approachable and tech-forward.

The brand identity carried across the product interfaces, the marketing site, and pitch materials. The tone was clear and supportive: a guide, not a gimmick.

I designed and built a lightweight Webflow marketing site that could evolve as the product did. It was easy to update as priorities shifted from demo signups to partner presentations.

The graphics were just as flexible: lightweight, dynamic, and designed to support the story without overshadowing the product.

None of the four directions moved forward. The product was still searching for traction, and soon after this work the company pivoted to a completely different idea that became Momentum AI. I don’t see the exploration as wasted, though. Each prototype was a real answer to the trust problem we set out to solve, and getting four of them in front of stakeholders quickly showed which bets were worth keeping. That speed and range is what let the team move on with confidence.

This project taught me that adaptability is as important as craft. The stakeholder’s needs shifted constantly, new priorities, new perspectives, sometimes a complete change in direction. Rather than resisting it, I learned to design in options. Four interface prototypes wasn’t a luxury; it was the only way to keep pace with a product that was still finding itself.

Working as the sole designer at a startup, while still learning the fundamentals in bootcamp, was intense. But it gave me something I couldn’t have gotten from coursework alone: real constraints, real stakeholders, real deadlines.

I would have pushed deeper into personalization: shopping history, saved preferences, visual search where you upload a photo and find similar items. I also wanted to test different AI personalities, friendlier, bolder, more consultative, to see which tone made users feel most supported.

The bigger question I didn’t get to explore: how do you build long-term trust with an AI shopping assistant? The first interaction gets users in the door. What keeps them coming back?