B. Speaks

Founder Aline Lin had built an early version of a lightweight user-feedback tool. She needed to validate the right wedge into the association market before committing to a rebuild. I led an end-to-end discovery program from scratch. I ran founder interviews to clarify the vision and constraints, two rounds of stakeholder interviews (internal team and external consultants who sell to associations), a competitive landscape review across enterprise platforms and indirect competitors, and structured workshops to converge on an ICP and MVP scope.

Associations have deep anecdotal knowledge of their members but almost no behavioral data to validate it. They do not want a research repository. They want push-based, specific, actionable recommendations they can hand to leadership. MARCOM teams at mid-tier trade and professional associations emerged as the best early adopters: under pressure to modernize, burned by past vendors, and operating in the 100 to 1,500 USD/month department-budget window.

The team shipped a focused MVP, a micro-tool that captures sentiment and friction at the point of contact, with AI-generated insight cards as the core unit of value. We defined a clear, defensible ICP and a narrow go-to-market wedge the team could use to target its first ten customer conversations.

Zintera was a Samsung-funded startup with an unproven neural network and a single bet: pairing human expertise with machine learning would beat either one alone. As Senior UX Researcher, I led customer discovery and field research alongside Pete Bouchard (Distinguished Engineer, IBM Watson Healthcare) and Dr. Syrus Nemat-Nasser (Stanford). We proposed, and I named, the Centaur Model of Artificial Intelligence: AI handles computation and pattern recognition; humans provide goal-setting, common-sense reasoning, and judgment.

The work started in the field. I shadowed thirteen nurse practitioners over two months as they audited complex hospital bills, capturing the artifacts they used, the decisions they made, and the data quality problems that slowed them down. I interviewed data scientists, infectious disease experts at Emory, and healthcare operators to map gaps between user mental models and the system's conceptual model. That gap determines whether AI is adopted or rejected.

The research translated directly into product strategy. We designed interfaces that treated trust as a first-class problem, gave human operators clear ways to correct the model and improve accuracy, and grounded use cases in the messy data realities of hospital billing. The first commercial Centaur deployment audited hospital bills for Kaiser Permanente in 2014. Solo human auditors averaged about 80% accuracy on this work and degraded across long shifts. The Centaur pairing held roughly 90% accuracy across full shifts without that fatigue curve, making the performance gain durable rather than peak.

Penn Medicine was rebuilding its provider-facing experience and needed to know whether the new navigation, content groupings, and labels matched users' mental models before committing to engineering. I led a three-phase research program: discovery, solutioning, and validation. Each phase used mixed methods to build evidence before launch.

In the solutioning phase, I ran moderated open card sorting with 18 participants and 50 cards. Sessions were conducted over remote video so I could capture reasoning, not just categorization. I then ran unmoderated closed card sorting in Maze with 20 participants, using categories that emerged from round one. High-agreement categories included Billing & Insurance, Make an Appointment, My Account, Find a Doctor, Research, Locations, and About the Company. Lower-agreement categories revealed overlapping mental models that needed reorganization.

Validation included moderated mobile prototype testing with five general healthcare consumers, unmoderated desktop testing with ten more, and a tree test with 20 participants across eight tasks. The work surfaced more than ten usability fixes before launch. These included a patient-stories module users did not trust, a hierarchy issue that buried specialties below dining options, and a Locations card pattern that performed better with icons. The result was a validated top-level navigation and an implementation-ready information architecture.

Regeneron's change control process, owned by QA but touching almost every major division, had become complex, error-prone, and hard to navigate. Leadership needed a way to scope a short-term project without losing sight of the longer-term strategy. I served as workshop facilitator for a two-day discovery engagement with nine cross-functional stakeholders.

The structure blended contextual research with collaborative innovation games from "Gamestorming" and "Innovation Games." We began by observing a live Change Control Committee meeting to ground the group in real behavior. Participants then worked through Business Process Mapping (storyboarding their narratives), Remembering the Future (defining success at December 2017 and 2018), Personas, Dinosaur Steps (segmenting the process into phases), Speedboat (visualizing pain points as anchors), and a round-robin business-requirements exercise, converging into affinity diagramming and dot voting.

The synthesis produced four prioritized themes: Ease of Use, Duplication & Alignment, Reporting, and Originating Activities Visibility. Each was linked to business impact and a future-state opportunity. The recommendation included a cross-functional team structure (Change Control, QRM, RA, Supply Chain Compliance, IT) and a phased approach that started with change control initiation and the directly impacted systems. This work fed directly into the roadmap for the next phase of the engagement.

Gigzilla was a mobile and desktop platform for skilled trade professionals. These included truck drivers, forklift operators, HVAC technicians, and others with certifications that do not fit neatly into LinkedIn. As cofounder and Manager of UX Research, I built the research practice while we built the product. The two grew together.

I conducted more than 200 remote and in-person user interviews. I facilitated bi-weekly focus groups with two to six participants who reviewed features and clickable prototypes. I also ran continuous usability studies to improve the information architecture and interaction design of the platform. Three insights from this discovery work proved pivotal. Trade workers needed credential portability. Generic job boards did not meet their needs. Community college workforce programs were a powerful distribution channel. Together, these insights moved the product to product-market fit.

I managed and mentored a small UX team. The team included three qualitative researchers, one designer, and one writer. I translated findings into personas, journey maps, and information architecture that the product was built on. I presented user insights and product strategy in concise, executive-ready formats. That work helped secure $5.5M in angel, pre-seed, and seed funding. The company exited after community college workforce programs adopted Gigzilla as a LinkedIn alternative.