AI-powered research platform
User-led AI - Led design for an AI research library used by 800,000+ researchers, shaped by interviews and testing.
At a glance
- Product
- Zendy, an AI research library with 120M+ papers and a built-in AI assistant for summaries, insights and recommendations
- My role
- Product Designer — set priorities with PMs and ML engineers; led AI brainstorming workshops to shape what to automate
- Scope
- Took the research platform and its AI features, including the domain-specific assistant, from concept to launch in 12+ countries in about a year.
The challenge
Zendy set out to make academic research affordable and accessible to everyone, especially students and researchers in places where library budgets are shrinking and papers sit behind paywalls. The library itself was huge, but users weren’t sticking around. They’d land, get lost, and leave, so bounce rates stayed high, subscriptions were weak, and few users came back.
A review of the platform, using Hotjar recordings, user interviews, and the issue backlog, showed why. Navigation trapped users, the UI was inconsistent, and there was no real brand identity or design system. Search returned a huge pool of material with very few ways to narrow it down. Meanwhile, competitors were already using AI for summaries, citations, and notes. Zendy needed to raise its game, starting from how students and academics actually do research and building the usability and automation around that.
My role
I led design and shaped the design strategy, working with the product manager and ML engineers on priorities and a phased delivery plan. That covered a full overhaul of the web and mobile experience, plus a new brand identity so the marketing website and the platform finally felt like one product.
To ground the work in real research habits, I ran interviews and workshops with university students and academics about how they use a research platform, what their scholarly workflow needs, and which features would make them stay. I also led AI brainstorming workshops to figure out which research tasks we could automate. Then I designed those features end to end, including taking Zendy’s domain-specific research assistant from concept to launch.
Quick wins first
Before any big redesign, I wanted Zendy to feel easier to use straight away. In my first few weeks, we used Hotjar and other usability testing tools to find where users were getting stuck, then cleared those bottlenecks one at a time. We smoothed out onboarding so first-time visitors could go straight into the platform and see what it could do, instead of getting stuck at login and sign-up. We moved the paywall to after users had seen the value, so it no longer blocked them before they’d had a chance to explore. We also fixed accessibility issues and added tooltips and cues so people always knew where they were heading. Each fix was small, but together they gave us early wins and real momentum.
Next, I took the initiative to map the critical user journeys, and turned them into a shared document the whole team used to plan new features. We weighed each feature by impact against effort, so we knew what would make the biggest difference, what would take longest, and what we could ship quickly with the people we had. The result was a roadmap the team understood and agreed on, built around real development effort rather than guesswork.
The solution
A library people could find their way around
With a collection this big, the design job was helping people move through it without getting lost. We rebuilt navigation and filtering and brought the whole UI onto one consistent design system. Every wireframe and journey came straight from pain points in interviews and usability tests.
Knowing what to open before opening it
Researchers were opening paper after paper just to check if it was relevant. Each search result now carried a short AI summary they could skim first, and their feed brought up new papers in their field without a separate hunt.
An assistant that answers across the literature
I took Zendy’s domain-specific research assistant from concept to launch. Researchers could ask a question in their own field and get an answer drawn from many papers, with references, which made literature reviews much faster.
Help inside a single paper
Once a paper was open, Reader AI helped researchers work through it. It summarised sections, highlighted key concepts, answered questions about that document, and supported citations, all without leaving the page.
The outcome
Post-launch interviews and usability testing showed users had much more confidence in the product. Feedback showed that AI summaries helped them find relevant content faster, and personalised recommendations stood out as a big win.
In that period we also launched in 12+ countries within about a year, tuning the experience for what each market needed as we expanded.
The company grew from about 300k to 800k users, with 120M+ research records.