QUALISIS
Designing AI support for thematic analysis while preserving researcher control
Project Overview
QualiSIS is my master’s thesis project exploring human–AI collaboration in qualitative data analysis.
Traditionally, analyzing qualitative data like interview transcripts is a highly manual, time-consuming process of reading, coding, and interpreting themes.
I designed and built QualiSIS to investigate how AI can speed up this process without replacing the researcher. The goal was to find a collaborative workflow that makes analysis more efficient while keeping humans in control of the final interpretation.
My role
As UX Researcher & Product Designer, I led the thesis project end-to-end, from defining the research questions and designing the AI-assisted workflow to building the prototype, conducting studies with 14 researchers, and translating findings into design improvements.
TRY THE LIVE PRODUCT
Product Walkthrough
A quick demonstration of how the QualiSIS prototype translates academic theory into a functional UI, focusing on traceability and evidence-based AI interactions.
Starting Point
QualiSIS started from a research pain point: while conducting thematic analysis, researchers struggled to find a reliable workflow; coding was manual and time-consuming, while fully automated AI tools reduced transparency and control. This inspired a more flexible way for researchers to analyze data efficiently, receive AI assistance, and maintain agency over the final results.
While analyzing qualitative data, I realized how difficult it was to balance AI speed with human control — and that became the starting point for this project.
The Process
Research
- Literature Review
- User Interviews
Synthesis
- Design Goals
- Core Features
Ideation
- Workflow Design
- Low-fi Prototype
Evaluation
- Usability Testing
- User Feedback
Outcomes
- Design Guidelines
- Key Takeaways
Interviewing Researchers
To validate my initial assumptions, I interviewed 14 qualitative researchers to understand their current thematic analysis workflow, the challenges they face, and their expectations for a digital tool. Afterwards, I synthesized the raw notes using an affinity map, which revealed 3 core problems.
Time-Consuming Coding Process
Researchers spend hours manually reading and highlighting transcripts. Without automation, the process of finding patterns across dozens of documents is tedious.
Loss of Context with Black-Box AI
When using fully automated tools, researchers feel disconnected from the data. They cannot trace how AI generated its summaries back to the original quotes.
Difficulty Managing Evolving Codebooks
As new themes emerge, managing and updating the codebook becomes messy. Keeping consistency across multiple researchers is challenging.
Codebook Analysis
Based on the interviews, I conducted a deep dive into how researchers construct and maintain their codebooks. Understanding this structure was critical for designing an AI assistant that augments rather than disrupts their established methods.
Building the Prototype
Unlike a traditional design process where I would mock everything up in Figma, I moved straight to building a functional interactive prototype to test the human-AI interaction dynamics in real time.
Coding with AI Agents
To rapidly develop the application, I collaborated with AI coding agents. This approach allowed me to build out complex features—such as text highlighting, dynamic codebook generation, and LLM integrations—much faster than traditional hand-coding, proving the value of AI in the development lifecycle as well as the research process.