AI is part of my practice.
Not just a topic I research.
I shape corporate products to be AI-friendly. I build tools to accelerate UX research. And I use it daily to sharpen how I understand users’ needs in AI.
The challenge
AI makes researchers nervous; the concern is legitimate. Not only can AI mislead, but it also makes it easier to get the answer we’re looking for rather than the one we actually need.
Knowing how to use AI responsibly when conducting research is a skill. But, knowing when to step away from it entirely is the more important one.
My approach
- Ground my knowledge in what LLMs actually do and don't do
- Create my own AI tools to learn where they fail
- Experiment with AI to strengthen my understanding
UXR Claude skill
Featured · Claude · Open source
5
websites tested · 200+ reads · 15 repo clones
Outcome
Open-source skill that the research community is reading and cloning.
I built a Claude skill to replace myself in conducting usability heuristic evaluations, then ran it across five websites to see what would hold up.
It turns out AI can do a lot. But it’s probabilistic. Early iterations confidently flagged problems that didn’t exist, especially against dynamic content the model couldn’t actually verify. So, the skill could generate some findings but it could not decide which findings were valid and worthwhile pursuing. That’s where the researcher’s work begins.
Read Part 1: Building the skill
Read Part 2: Five Websites, One Claude Skill, and the Thing Prompts Can’t Fix
Persona automation
Personas · Python
In use today
across CIHI departments & marketing analytics
Outcome
Static personas turned into a system that teams act on.
I turned personas into a rule-based system to track audiences.
Personas sitting in a research report don’t change how teams make decisions. I worked with researchers and subject matter experts at the Canadian Institute for Health Information (CIHI) to define persona characteristics, then built rule-based classification systems that operationalized them across departments. CIHI’s marketing department adopted the system to segment clients systematically, enabling digital analytics that map signed-in visitors to the persona framework. Teams still use the outputs today.
Pre-dating current AI tools, my program was designed so machine learning could later replace more rudimentary automations. This meant I focused on the key goals of the automation processes without focusing on technical builds. Ultimately, landing on validated data structures that future AI could easily plug into.
Requirements discovery
Discovery · UX for APIs
3
generative studies — survey, interviews, concept testing
Outcome
Requirements that make AI actually work for users downstream.
Building a clear picture of the problem space before expensive product directions get set.
Early product decisions made without a clear picture of the problem space are expensive to undo. I help teams build that picture before the direction is set.
I work with technical leads to translate user insights into requirements for data, metadata, and APIs. Recent work led three generative studies (survey, interviews, concept testing) to inform requirements for a product that needs to support both human use and automated system access.
This user-centred collaboration is where research stops being a report and starts being implemented. These recommendations are what make AI actually work for users downstream.
What’s next
I’m building toward work that sits at the intersection of three threads I’ve been pulling at.
Practice Development
Designing a UX research repository that connects what researchers produce to what teams actually use when making product decisions. Currently in planning, with case studies to come.
AI Fluency
Deepening AI fluency through structured learning, including Anthropic’s AI Fluency Framework.
Responsible AI
Exploring how AI can build a stronger sense of duty to the people being researched, not just speed up research.