Ventrilo.aiRole: UX Research Intern, lead researcherSep 2025 - present
Building a research practice at an AI startup
I lead end-to-end research for an AI browser extension, from scoping questions and recruiting to sessions, synthesis, and share-outs teams act on.
10research cycles led end to end
10+core design changes driven
0 → 70+participant pool built
What I do
Run usability tests, in-depth interviews, heuristic reviews, and competitive research
Built the team's research repository, a shared source of insight for design, engineering, and leadership
Analyze usage and retention data to understand what separates engaged users from those who churn
What changed
Findings reshaped onboarding and core flows, including how people save and find content
Surfaced privacy concerns with third-party integrations, leading to transparent data access
A heuristic review added error-prevention safeguards, like confirming before deleting
Internal materials are confidential, so this case is described in words only.
CareAXLQualitative study for a healthtech incubatorSpring 2025Role: Qualitative Researcher (solo)
What makes care at home feel safe?
A healthtech incubator was building a platform to bring care into homes. I studied what makes that feel safe and trustworthy, who people trust with health decisions, and how they want to book.
5participants across 4 interview sessions
3research questions
2personas, from risk-averse to risk-tolerant
My role
Everything, end to end: screening criteria, a 75-minute interview protocol, remote interviews, thematic analysis, personas, and the report. Done in 5 weeks with no budget.
Key findings
Trust comes from communication, quality care, clear visit notes, and feeling understood by the provider
Insurance and cost decide which provider people choose, but have no bearing on trust
Most people trust family and friends for health decisions over clinicians
People book by phone today but want online booking with text or email confirmations
Trust model: the same four factors either build trust or break it, depending on how they show up.What drives provider choice, from most to least decisive.Personas mapped on a risk-averse to risk-tolerant scale.
WellVersed CareIntelligent User Interfaces (HF760), BentleySpring 2026Team of 3Role: UX Researcher & Designer
An AI plugin that makes medical notes understandable
An AI add-on for patient portals that turns doctors' notes into plain language, translates them, and answers follow-up questions from peer-reviewed sources, built for patients the system often underserves.
9ethical risks mapped, each with a mitigation
2think-aloud usability sessions
4design changes from testing
My role
Co-designed the concept and owned the end-to-end desktop flow (sign-in, consent, setup, chatbot), plus the ethics and fairness framing. Built the working prototype in Lovable with a live retrieval API (ZeroEntropy) searching 60 million peer-reviewed sources, so answers stay grounded in vetted evidence.
What we learned
Neither participant read consent unprompted, so consent had to be designed, not assumed
Personalization that stayed on after being switched off felt like a privacy breach, so we reworked data controls
AI summaries need clear AI labels; one participant didn't realize the summary was machine-written
Ethics plan covered bias, unequal access (multilingual, low-end devices), privacy, and informed consent
System architecture: consent comes before any data sharing, and every answer passes through plain-language and translation steps. Early version, before the team switched to the ZeroEntropy API.Consent screen: each kind of data use is a separate, explicit opt-in.PlainNotes: the original note, a plain-language summary with sourced references, and the AI chat side by side. Sample patient data.
Prevent BlindnessMeasuring the User Experience (HF751), BentleySpring 2026Team of 4Role: UX Researcher
Redesigning how people find eye-health information
The "Your Sight" menu on PreventBlindness.org had 76 links in about 20 top-level items. A card sort showed how people group the content, then a tree test compared the new structure to the old one.
20 · 61card-sort participants · cards
40tree-test participants, old vs new
25% → 75%rated the menu intuitive: old content grouping structure vs. new
My role
Wrote the card-sort method, sharpened the survey questions (removing leading and double-barrelled items), and built the Insurance and Eye Exams menu branches. The team analyzed results in R and JASP.
Results
People group eye-health content by their own needs (insurance, exams, conditions, aging, safety), not by the site's labels
The new menu felt far more intuitive: 75% of participants rated it intuitive, up from 25% for the old one
People got lost less and found answers more often; success on the macular degeneration coverage task rose from 35% to 60%
Cutting about 20 top-level items to 6 reduced the "too many options" overwhelm
Not every change helped: deeper nesting slowed the eclipse task and vague labels like "Resources" still misled, so both were flagged for the next round
Card-sort dendrogram: 61 cards clustered into the 9 categories that became the new menu.Tree-test results by task. Blue: new menu better. Orange: worse. *** marks a significant result.How intuitive the menu felt: new (Experiment) vs current (Benchmark).
kohlpharmaGlobal Health Case Challenge, Scheer School × BentleyNov 2025Team of 4Role: Market Researcher & Strategic Analyst
Where should a pharma importer grow next?
A German parallel-import pharma company asked which EU markets beyond Germany and Austria offer real potential. We built a repeatable scoring model to answer it.
7 · 25dimensions · sub-criteria in the model
27 → 4EU markets narrowed to finalists
96 vs 76Germany vs Austria benchmark score (of 120)
My role
Made the framework measurable by replacing vague criteria with measurable ones, proposed Germany and Austria as a validation baseline, co-led the first country screen, and coordinated the team across time zones.
Approach
Multi-criteria analysis: each sub-criterion scored 1-5, weighted, and tiered
Public data from OECD, Eurostat, EFPIA, PPRI, and WHO
Validated on the client's existing markets before scoring new ones
Finalists: Sweden, the Netherlands, France, and Spain
Screening funnel. Redrawn from the team's framework.Model validation on existing markets, as a share of each dimension's maximum. Redrawn from the team's scores.
Global Business Experience: JapanBentley GBE 790Winter 2025-26Role: Independent Researcher (solo)
Does traditional practice shape mental-health care in Japan?
How far do Kampo (traditional herbal medicine) and Ma (meaningful space and pause) shape modern Japanese mental-health care? I tested my assumptions in the field.
間
Method
A review of about 20 sources, then multi-site fieldwork at Kampo pharmacies, temples, a tea ceremony, and a ryokan, plus talks with Kampo practitioners and an architect.
What I found
None of my three hypotheses held as stated, which was the useful part. It raised a new question: the gap between Japan's calm-oriented culture and its mental-health realities, like stigma and limited insurance coverage for therapy.
Expected: Kampo is a first resort for mental healthFound: Western medicine leads; Kampo often manages side effects
Expected: care is deliberately paced around MaFound: pacing comes from the wider culture, not patient-experience design
Expected: Ma is a design rule for care spacesFound: it is a cultural sensitivity to atmosphere, not an explicit rule