Research that
moves products
forward.

The Challenge

The client's IDE was incompatible with unmoderated testing platforms — a limitation discovered after the project began. Their legal team also prohibited recruiting existing users, forcing a pivot in the study population. And an open-ended IDE poses an inherently difficult task-definition problem: what counts as a representative workflow?

My Approach

• Pivoted to moderated testing not in the original scope — trained junior researchers on a facilitation protocol I developed to ensure consistency across 40 sessions
• Reframed the population: shifted from existing users to professional developers exploring low-code for the first time, aligning with the product's growth-phase goals
• Ran a collaborative workshop with the client team to surface the top developer workflows for the first hour of use — used these to define representative tasks
• Coordinated a zero-state test environment with the technical team so every participant started with the same conditions
• Led qualitative pilot (n=10) to surface mental models and pain points, which also served as a live training ground for junior researchers before the quant phase

Research Process

• Qualitative pilot study: 10 participants, mental model elicitation + think-aloud
• Quantitative study: 40 participants, task success, time-on-task, error rates, SUS scores, desirability word selection
• Attitudinal survey: competitive benchmarking against comparable platforms
• Statistical analysis in R: subgroup analysis across developer experience levels, tool preferences, attitudinal clusters

Outcomes & Impact

• Established baseline UX metrics and a repeatable benchmarking framework for tracking future improvements
• Delivered a nuanced competitive analysis triangulating SUS scores, behavioral data, and attitudinal responses
• Roadmap prioritization directly influenced: redesigned feedback system and clearer terminology shipped
• VP of Product: "This is an exceptional piece of work. It will define our roadmap for the next two years."
• Research operations benefit: refined junior researcher training process now used across NN/g client engagements

The Challenge

Three simultaneous hard problems: (1) A domain so specialized that observing real work meant understanding ICD coding at a clinical level. (2) PHI constraints that made conventional contextual research legally off-limits. (3) An engineering-led product team that didn't believe UX would tell them anything useful. Stakeholder buy-in was far from guaranteed going in.

My Approach

• Built domain expertise first: invested heavily in relationships with ICD coding experts and co-created the research approach with them rather than around them
• PHI workaround: used contextual inquiry for real-world observation, then switched to carefully constructed simulated data for iterative prototype testing — maintaining research validity while staying compliant
• Turned skeptics into champions: brought engineers into the research process directly, making them observers and collaborators rather than audience members
• Built a one-day research bootcamp for the product team — they left able to participate meaningfully in future studies
• Mentored a junior researcher through their first end-to-end client engagement

Research Process

• Stakeholder interviews to scope the project and map organizational priorities
• Contextual inquiry: observed real coders working in their actual environment
• Heuristic evaluation of the existing interface to catalog friction points
• Iterative prototype testing using simulated PHI-compliant data across multiple design rounds
• Workshop facilitation to align the team on findings and co-prioritize solutions

Outcomes & Impact

• New visual design system replacing a 20-year-old interface — shipped to 18 registry locations nationally
• Redesigned workflows reduced reliance on working memory and measurably reduced error rates
• Post-launch feedback showed significant improvements in user satisfaction across the registry network
• Produced a 3–5 year strategic roadmap for continued UX investment
• Cultural shift: a previously skeptical engineering org now runs iterative UX research as standard practice

The Challenge

Generative AI poses fundamental research challenges: output is probabilistic and variable, making consistent stimuli difficult to construct. User mental models of AI are often wildly inaccurate. And much of the most consequential design territory — explainability features, confidence indicators, failure-mode transparency — involves design patterns that don't yet widely exist. How do you study user reactions to interfaces that haven't been built yet?

My Approach

• Developed controlled prompting scenarios to give participants consistent experiences despite the probabilistic nature of AI output
• Combined mental model elicitation with think-aloud usability sessions to capture both beliefs and behavior
• Designed forward-looking research protocols that test reactions to hypothetical explainability features and confidence indicators
• Planned longitudinal diary phase to capture how trust in AI features changes over real-world use across multi-week integration
• Co-researcher collaboration with structured alignment checkpoints to maintain consistency

Research Design

• Phase 1 (current): Usability testing and mental model elicitation — foundational trust beliefs and prompt-generation strategies
• Phase 2 (planned): Longitudinal diary study — how users integrate AI features into real information-seeking over time
• Methodological contribution: sample-size guidelines for AI UX research, handling output variability, designing stimuli for non-deterministic systems

Anticipated Outputs

• Design principles for interaction patterns that foster appropriate trust calibration
• Mental model taxonomy: common user beliefs about AI that designers need to design for (or against)
• Research methodology contributions: reusable frameworks for AI UX studies
• Publication series building on my existing NN/g AI writing (hallucinations, explainability, agentic interfaces)