How I think, make, and help teams move forward.

The problem was not speech. It was knowing what to say.

A graphical interface usually makes at least some actions visible: a button suggests a click, a menu hints at choices. In a voice interface, possible actions can remain hidden until someone already knows the vocabulary. For people who rely on hands-free interaction, that lack of discoverability is more than an inconvenience—it can determine whether a task is possible.

Our research focused on multi-step mobile tasks, where a user must move through changing interface states. We treated voice as an interaction design problem involving feedback, learning, and expectations—not simply a speech-recognition feature.

Use evaluation to reveal the interaction model.

I led the research and design process. An initial ten usability sessions with nine participants exposed problems with command discovery, interpreting the system's listening behavior, and managing on-screen guidance. These findings informed iterations to numbered overlays, contextual tutor messages, and a “What Can I Say?” menu. A revised prototype was then evaluated in five additional sessions.

Accessibility is also a question of what the system makes knowable.

The familiar design principle of making actions visible becomes difficult when an interface is operated by speech. Voice commands can be invisible, and a user may not know whether a failed attempt reflects recognition, unavailable functionality, or an incorrect mental model. The interface therefore has a responsibility to communicate its boundaries and offer recoverable paths forward.

This is one reason the project remains relevant to my later work on AI-mediated interaction. Systems that accept open-ended language can feel flexible while concealing their actual capabilities. Good interaction design needs to help people form calibrated expectations, rather than merely invite them to try.

This connection to contemporary AI interfaces is a present-day interpretation of the design implications, not a claim that the 2016 study evaluated generative AI or agentic systems.

Designing for agency means designing for discovery and recovery.

The work produced evaluated concepts and a clearer account of the trade-offs between discoverability, learning, and interface complexity. It did not establish long-term use or deployment at scale. Its enduring lesson for me is that an alternative input method is only accessible when people can understand how to use it and how the system responds.

What if the request for better information is the wrong starting point?

Urban sensing systems are often installed in public spaces where residents encounter their consequences but may have little involvement in decisions about their use. Our city collaborators began with a plausible communication question: what should physical signage tell people about sensors?

That question already defined transparency as an information-delivery problem. Our research-through-design approach deliberately kept the premise open. A sign could be a candidate interface, but it could also become a way for residents to articulate what was missing from the city's relationship with them.

Making a sign became a way to ask a different question.

I led the workshop research design, facilitation, partnership coordination, synthesis, and writing. In three workshops, 40 NYC residents worked with physical materials to create signs and discuss what a person should know or be able to do when encountering urban sensing technology.

These artifacts were not treated only as competing layouts to be scored against an existing brief. They gave participants a concrete medium through which to express concerns, speculate about alternative civic relationships, and challenge what transparency should accomplish.

Transparency is not the same as accountability.

Disclosure is valuable, but information alone does not determine whether people can challenge an institution's decisions. A sign may tell residents that a sensor exists without addressing who authorized it, who benefits, how data is used, or what avenues for contestation exist.

Our analysis therefore treats transparency as a social and political relation, not only an interface property. The project suggests that participatory design can help residents question whether a technology should be deployed, how it should be governed, and what responsibilities its operators carry.

This is the academic contribution I would foreground on a research-oriented website: the work is simultaneously an exploration of signage as an interaction medium and a critique of the assumption that greater disclosure automatically produces public trust.

These are interpretations of the participatory inquiry, not evidence that any particular governance reform was implemented.

Sometimes the most valuable prototype changes the brief.

Leading the project required balancing a city partner's practical question with a research process open to more fundamental challenges. Rather than promise a definitive signage system, we produced a set of findings about the limits and opportunities of transparency and the role of participatory design in reframing civic technology.

The resulting CHI 2024 paper received a Best Paper Honorable Mention. For my design practice, the lasting lesson is that prototyping and participation can produce knowledge that a conventional requirements process might never solicit.

Designing from inside an institution's everyday work.

The project grew from design ethnography with Atlanta's Mayor's Office of Immigrant Affairs. Rather than beginning with direct access to residents, the inquiry examined how staff built relationships with immigrant communities and helped people navigate housing-code concerns. That distinction matters: office staff were the principal design collaborators, and the research should not be represented as a resident co-design process.

I led the fieldwork and early UX framing, produced exploratory sketches, and coordinated with a designer and programmer. The goal was to turn situated knowledge about the office's practices into a coherent service concept without treating trust as a transaction that software could simply automate.

The sketches reveal how the service took shape.

Contemporaneous notes from a November 2019 design review show how paper prototypes supported conversations about information, institutional roles, and process complexity. The housing-rights guide was revised to be less text-heavy and more visual. A larger illustration of a house made housing-code issues easier to point to, while representations of residents, landlords, and enforcement officers helped clarify their different roles.

The design review was itself a research setting. Staff responded to artifacts, described how they handled complaints, and exposed assumptions about who would take responsibility for each step. These conversations shaped the design while also revealing the institutional politics behind apparently straightforward UX decisions.

Removing steps was not the same as resolving the problem.

The tension between administrative efficiency and relational trust was the finding.

One reading of the design is that it makes a difficult municipal process more usable. A second, more critical reading asks what happens when public institutions make services efficient by shifting responsibility onto residents. The research describes this tension through the idea of trust as distance: social, institutional, and power relations shape whether a service can be trusted, and those relations cannot be solved by interface polish alone.

My own fieldnotes are candid about moments when I advocated streamlining steps even as I questioned whether this might diminish the office's relational work. That tension is part of the design research, not something to edit out of the story. The same tools that could withdraw staff from direct mediation could also make it easier to maintain contact and follow up.

As a designer and researcher, I take this project as a reminder that a good workflow is not necessarily a good institutional arrangement. A service needs to be evaluated for what it enables, whose labor it changes, and what responsibilities it redistributes.

Evidence: original November 15, 2019 design-review notes and Corbett & Le Dantec, CHI 2021. Staff names and personally identifying details from fieldnotes are intentionally excluded from the public narrative.