Designing awareness without demanding attention
Mezzo began as a real-time volume display. Working software showed that useful feedback could still fail if people had to watch it, manage it, or question whether it applied. I reframed it as an ambient companion designed to support awareness through a glance.
Private, ambient feedback. Mezzo measures volume locally and provides a subtle signal relative to a preferred range. It never records or transcribes speech.
From a real-time volume display to
a trustworthy ambient companion
Awareness only helps when people can trust it without giving it their attention.
Working software revealed two hidden costs: the attention required to monitor the signal and the effort required to decide whether it could be trusted.
When private self-correction failed in shared spaces, the cost shifted to everyone nearby
As return-to-office policies brought more people back into shared workplaces, more calls moved to desks, open areas, and other spaces not designed for focused conversation.
For some people, noise-canceling headphones weakened the natural feedback they relied on to regulate their own voice. In a shared environment, someone could get louder without realizing it, turning a private awareness gap into a disruption other people had to manage.
A call happens in a shared space
The speaker is surrounded by people trying to do their own work.
Headphones weaken feedback, and volume drifts
The speaker hears less of their own voice and can become louder without realizing it.
The consequence becomes shared
- Nearby coworkers lose focus or struggle to continue working.
- Someone else must tolerate it, move, signal indirectly, or speak up.
I had experienced this awareness gap myself and could see the value of a private signal that arrived before anyone else had to react.
Could software restore private awareness before other people had to manage the consequences?
Each round changed what the product needed to solve
The work moved through three stages: identifying the audience and emotional stakes, testing whether the concept and tone made sense, and evaluating a deliberately designed experience in real meetings. Each round narrowed the product before the current MVP was built.
The problem mattered deeply to a specific audience, and public correction carried a personal cost
I interviewed five people about speaking volume in shared workplaces. Some already worried about disrupting coworkers, while others rarely considered their own volume. The interviews narrowed Mezzo’s audience to people who actively wanted private awareness rather than a workplace correction tool imposed on everyone.
They also revealed a cost I had not anticipated. Being asked to quiet down could make someone self-conscious, reduce their confidence, and change how they participated for the rest of the meeting, especially when their volume came from enthusiasm or a naturally strong voice.
What changed Public correction did more than solve a volume problem; it made people self-conscious and less effective.
People understood the concept, and the feedback felt neutral rather than corrective
I showed five participants a recorded demonstration of a working prototype built in Claude Code, with limited usability refinements before testing. All five correctly understood that it compared their speaking volume with a target range and provided real-time feedback.
The tone also landed as intended. Participants described the feedback as fair, clear, neutral, and not distracting.
The study surfaced two unresolved product questions. Three participants independently asked where the indicator would live during screen sharing. Two questioned whether people who were unaware of their volume would choose to use the tool at all. Several suggested integrating it into Teams, Zoom, the browser, or the operating system rather than relying on a standalone tab.
What changed The signal felt neutral, but participants questioned where it would live during a real meeting.
Real use showed that useful feedback could still fail when it competed for attention or was forgotten entirely
Two participants used a redesigned live prototype across three meetings. The experience included ambient states, revised language, a calmer visual system, and an emerging Companion workflow. This round evaluated not only whether the product worked, but whether its visuals, terminology, and tone felt helpful rather than corrective.
The states were understandable and the tone felt neutral. But real use exposed issues a recorded demonstration could not: the signal competed with the meeting, people could forget to launch it, setup and monitoring were split across surfaces, and the reference point did not feel equally trustworthy across environments.
What changed Real use exposed two additional product problems: divided attention during meetings and forgotten activation between them.
The question changed from displaying volume to designing a system people could notice, begin, and trust
Round 3 showed that a neutral signal was possible, but the signal alone was not the product. The MVP also had to reduce activation effort, stay peripheral during meetings, clarify where tasks belonged, and provide a credible reference across contexts.
How might we tell someone when they are speaking too loudly?
How might we restore private awareness before disruption becomes social, without adding a new task to manage?
Three product decisions reshaped what the user sees, where the experience lives, and what the system remembers
Round 3 translated the reframed problem into three changes to the product model: a calmer signal, a reference model that delivers value before asking users to save, and one operational home for the session.
Feedback became a neutral, ambient signal
The system continues measuring microphone level precisely, but the interface exposes four recognizable states: Quiet, Within Range, Approaching End of Range, and Outside Range. The same tone carries into the Ready state, which acknowledges the user’s intention without evaluating how the meeting went.
Participants understood the signal and described its tone as neutral and non-distracting, while live use showed that detailed feedback competed with the conversation.
A simpler signal reduces interpretation work, but it still has to respond clearly enough for people to connect the feedback with their speech. Supportive language must also avoid implying achievement, performance, or a successful meeting outcome.
Calibration became useful before it became permanent
Calibration now creates a Temporary profile that can be used immediately and reused while the Companion remains open. Saving is optional and turns that profile into a persistent Speaking profile.
- One universal default offered the fastest start but the weakest contextual trust.
- Always calibrating improved fit but added repeated setup burden.
- Requiring every calibration to be saved introduced commitment before value.
- Saved profiles alone made return use fast but implied that an old calibration was still correct.
Supporting both Temporary profile and Speaking profile adds conceptual complexity, but it preserves useful calibration work, avoids false precision across environments, and prevents repeated save-or-discard decisions.
A saved profile changes only through deliberate action. Recalibrate replaces the range from within that profile; Update assigns an already completed Temporary profile to an existing profile after explaining what will be overwritten.
The Companion became the session’s operational home
Selecting a Speaking profile or a Temporary profile, starting monitoring, receiving feedback, stopping, returning to Ready, and beginning another session now happen inside one small floating Companion. The web experience was reduced to launch and docking responsibilities.
- A Zoom add-in could improve meeting proximity, but would make Mezzo platform-specific.
- A web app enabled browser-based picture-in-picture, but depended on browser lifecycle.
- A browser extension created a persistent entry point, but remained a launcher rather than true meeting integration.
Picture-in-picture keeps feedback close to the conversation across meeting platforms, but activation remains weaker than native integration. The Ready state can make the return feel continuous and supportive, but it cannot become a performance summary or compete with Start monitoring.
The extension provides a persistent route back into Mezzo. After monitoring stops, the Ready state acknowledges the user’s intention and returns the primary action to Start monitoring.
Together, these decisions moved Mezzo from a volume display toward a system that could be noticed peripherally, trusted contextually, and restarted predictably.
The next question is whether Mezzo works across repeated real meetings
The first three rounds shaped the product model. The next study will test whether people remember to use Mezzo, notice it without watching it, trust its reference, and adjust without becoming more self-conscious.
- The problem matters most to people who already want private awareness.
- Participants understood the ambient states, and the tone felt neutral rather than corrective.
- Attention, placement, activation, and reference trust determine whether the feedback is useful.
- Local processing supports the experience without recording or transcribing speech.
The next study must test both usefulness and its cost
Mezzo should not be judged only by whether speaking volume changes. The study must also show that people remember to use it, notice the signal without watching it, trust its reference, and continue participating naturally.
The real design problem was attention, not feedback quality
Working software showed me that the more important design problem was everything the user had to do to receive and trust the signal.
Early versions asked for the same amount of attention no matter what was happening, a visible number or bar the user had to keep checking. Round 3 testing showed that constant attention was itself the problem: people forgot to launch it, lost track of it mid-meeting, or tuned it out because it competed with the conversation. The feedback became useful once it stopped asking for attention at a flat rate and started asking for more only as the risk to others grew.