Larn AI

Improving Usability of Maven Central and increasing awareness of Sonatype Nexus Respository

User Research

User Testing

Wireframes

Hi-Fidelity

My role

User Research, Wireframing, Accessibility & Visual Design.

Platform

Desktop, Mobile Responsive

Duration

2 months

Overview

Larn is an AI-powered research assistant designed to help academic and industry researchers streamline their research process including literature review, data collection and insight generation. This product is currently in beta.

I led the product’s redesign, working closely with the CEO and engineering team to rethink the experience from the ground up with the aim of transforming Larn into a more focused, intuitive tool for researchers doing meaningful work.

Why It Mattered

What we wanted to achieve

The goal was to reduce the time and cognitive load spent on confirming the relevancy of a research article thus enabling researchers to focus on high-level thinking and innovation.

What is currently happening

With limited resources, our interviews focused solely on academic researchers and students, conducted through surveys and usage monitoring.
From earlier research we also know their primary goal is to publish research papers often as a means to secure future contracts.
‍
The global research software market (including tools for data analysis, lab management, and collaboration) is projected to grow steadily, with annual growth rates between 8–12%, depending on the segment.
‍
‍AI-powered tools for literature review, citation management, and research workflow (like Scite, ResearchRabbit, Elicit, Consensus AI, Anara) have raised significant funding and are seeing increased adoption among academics and corporate R&D teams.

Softwaere Research Market

AI - Powered Tools

What we knew about our users

Type 1: Casual users: These users tried Larn once or twice and never returned. Typically students with no ongoing research needs.
Type 2: Occasional users: They used Larn 2–3 times bi-weekly, often for long sessions. These were typically students who occasionally needed research support.
Type 3: Core users: They used Larn consistently, 4–5 times in a row at short intervals. These were typically involved in primary research fields.

In the early stages, our approach was largely experimental. Instead of creating rigid personas, we categorized users based on how they used Larn.

We analyzed the workflow of Larn’s core users (Type 3) and observed a clear pattern: they used the chat to quickly judge a document’s relevance. Irrelevant documents were usually dismissed within 3–4 prompts, while the relevant ones led to 10–12 prompts.

User Types

What we struggled with

I worked closely with the CEO to recreate what we wanted out of Larn. The first set of wireframes made the user focus on the document rather than the chat interface as shown in the original Larn screens.

We weren’t fully confident in our decision to encapsulate the document within the chat interface, especially since competitors who had been in the space longer used side-by-side layouts.

Although our goals differed, it raised doubts about whether we were on the right path.

Original Larn UI - Chat Interface

Original Larn UI - Article Interface

Wireframe - Article Interface

Wireframe - Project (Grid)

Wireframe - Project (List)

Design

The first set of wireframes made the user focus on the document rather than the chat interface as shown in the original Larn screens.

After initial feedback, we found out users felt over stimulated with the large amount of text in view. This feedback helped us getting closer to our goal of reducing the time spent to check the relevancy of an article to the user's research.

The interface has been enhanced with two key additions including a project workspace for organizing multiple research documents, and a campaign tool that evaluates the relevance of many documents against user-defined criteria using a scale of 1-5 to infer relevance.

Style Guide

Dashboard

Larn - Onboarding

Larn - Upload Article

Larn - Project (No files)

Larn - Project (Grid /List)

Feedback

After testing the updated workflow with users, we found they appreciated the workflow provided using the campaign tool. We observed users placed more emphasis on certain relevance metrics over the other while also having limited understanding on some metric options.

These issues were handled with tooltip options to provide more context to a question as well as revising the metrics to make the workflow shorter.

Larn - Create Campaign

Larn View Article - Select Metrics

impact

We measured impact by comparing the amount of time users spent on checking the relevancy of multiple documents (up to 5 documents). We observed withe campaign tool users would spend significantly more time on the first document compared to a user just using the chatbot. However with consecutive tries with different documents, the time is significantly cut short.

The changes led to an overall average faster time in reducing the amount of time in which a user spends on checking the relevancy of a document by 496.01%. (approximately 500%).

COnclusion

Redesigning Larn was a process of learning through iteration and real user behavior. By focusing on our most engaged users, we built features that aligned with their research needs, simplified their workflows, and helped researchers judge each paper with clarity.

While there’s still more to explore, these early changes positioned Larn as a smarter, more focused tool for researchers doing meaningful work.

Previous : Sonatype Home.