I research and design how people interact with AI systems.

AI tutors are being adopted faster than anyone has measured their effect on the learner. I wanted to know whether adapting the tutor's persona to respond to the user's preferred learning style would change how users learn math. We wanted to know whether there was a change in the user's subjective, lived, and measured experience when comparing a personalized AI tutor with a non-personalized one.

Through prompt engineering, I built tutors that adapted their delivery to each learner: more concrete and example-driven for some, more abstract and pattern-focused for others, mapped to Felder-Silverman learner profiles and written entirely into the system prompt. That choice matters for industry: Manipulating the system prompt costs nothing. Forty participants worked through a fully user-led AI-tutored math session on Venn diagrams, logarithms, and parabolas followed by a post-tutoring quiz. We recorded interaction logs, subjective measures (using validated scales), pupillometry, and EEG.

The quiz scores did not see any change across treatment conditions; however, the physiological data did see some significant results. Pupil dilation during tutoring was significantly lower with the personalized tutor, suggesting that learners processed the same material with lower cognitive load, as corroborated by the EEG results. They also engaged more, skipping fewer quiz questions and writing out more notation while working with the AI. Although test scores (outcome measures) remained untouched, the underlying interaction experience may have been influenced by the personalization of tutoring. For anyone building EdTech: a more personalized experience can make a difference, and if you only measure outcomes, you miss most of what happens (the cognitive efficiency gains). Our work was presented and published at the Florida AI Research Society's (FLAIRS) annual Conference.

ESDC employees received access to Microsoft Copilot, but not everyone knew how to use Gen AI tools confidently. My team wanted me to find ways to improve their own outputs using AI and ensure its responsible use. The Workplace Mental Health (WMH) team also wanted to make AI that could help public servants locate wellness resources much faster. With the help of the innovation team, we trialled a RAG-AI system.

I fine-tuned and tested a retrieval-augmented generation (RAG) system intended for internal use to improve findability of wellness resources. I designed personalized co-pilot agents to increase productivity and wrote AI training resources and reference guides to help colleagues use the tools. I consulted on AI implementation strategy aligned with the 13 psychological factors of a healthy workplace at ESDC action group meetings.

I built a toolkit of resources and reference guides to democratize AI knowledge and provide a starting point for using agents. Examples and use cases in the toolkit were framed around real tasks. We did not proceed with the RAG-AI project due to resource constraints.

The City of Calgary digital team had tested its digital services for years, but previous testers were not the people most affected when those services failed. Newcomers, older adults, and people with disabilities using assistive technology were largely absent from the design loop. Our job at IncluCity was to support the city's efforts to recruit underrepresented groups for testing.

As the program manager, I led recruitment, project management, and volunteer coordination across roughly 20 researchers between the two teams working simultaneously on this project, with a target of four equity groups and eight participants each. Recruiting participants with disabilities was difficult: none of our 600+ existing testers met the full criteria. I put three changes in place: a referral program with incentives that opened snowball recruitment, removal of a screening rule that filtered out eligible people, and high-touch onboarding for testers who weren't answering email.

The pilot mapped 18 distinct equity challenges across seven services. Three city service teams adopted the findings (Recreation, myID, Parks), a follow-up study on people with disabilities (group 5) was approved, and mobile-plus-desktop testing for the same audience became standard.

Beneva, one of Canada's largest mutual insurers, wanted to know whether real users could complete home and car quotes as smoothly as competitors advertised. They also wanted to explore why users drop off before completing a quote.

I was the study design and quantitative research specialist on this research team. We recruited 12 participants who had shopped for insurance in the past year. We evaluated both home and auto flows by separating them into two parts, making four task scenarios. We ran a mixed-methods observational study measuring scale items such as satisfaction and behavioural and system measures such as effectiveness and efficiency, and benchmarked these against competitors. Each section finished with a qualitative interview.

Six of twelve participants failed to complete the home-quote flow. We delivered eight prioritized fixes with severity ratings the product team could act on. I cannot openly discuss the results of this study; however, I can refer broadly to the methods and outcomes in an interview if contacted.