Field Guide
Designed the digital field data collection system for a 700-household NIH-funded cohort study, turning fragile paper infrastructure into the operational backbone for a landmark study in dental disease genetics.
High-impact science.
Fragile infrastructure.
The study was designed to be landmark science: a multi-site research effort tracking families across years in rural Appalachia, built on longitudinal precision and patterns that only emerge over time. But the operational infrastructure underneath it was built for a different era.
At each clinic visit, participants completed multi-page paper surveys by hand. Completed bubble sheets were physically transported to a central location, then manually entered into the research database — introducing transcription errors at every transfer point. There was no centralized system at the point of care. Researchers had no visibility into the data during sessions. By the time errors surfaced, the participant had already left.
Data from each session took weeks to become available for analysis. For a study that depended on longitudinal precision, that lag wasn't just inconvenient: it was a structural threat to the research itself.
The team said literacy.
The data said overload.
Paper surveys and manual workflows were creating random answer patterns beginning at the 34% completion mark, hours-long sessions, and families leaving before receiving the care they were promised. The research team assumed the issue was reading ability. That wasn't the full picture.
A trusted guide,
not a digital form.
We designed a voice-enabled, tablet-based Field Guide that transformed dense research protocol into a structured, human-centered workflow. The core design principle: every system decision must either build trust or keep a promise. If it did neither, we cut it.
From deterministic capture
to agentic orchestration.
The original system replaced paper with a structured tablet workflow. A reimagined v2 explores how longitudinal research infrastructure evolves when agents can detect fatigue, adapt phrasing, reconcile contradictions, and coordinate multi-actor flow in real time.
The system,
running.
The V2 system doesn't just capture data. It reasons about what it's seeing in real time — detecting fatigue, adapting its approach, and surfacing decisions for human review before anything is acted on.
