An AI-assisted wearable system that gives people with social anxiety discreet, dignified permission to leave before they reach their limit.
Watch Wizard of Oz DemoOverview
People with social anxiety often can't decide to leave when they're overwhelmed — not because they don't want to, but because social pressure outweighs their internal discomfort.
SoftSignal is an AI-assisted system that monitors physiological stress signals via a smartwatch — heart rate variability, tension patterns — and delivers a discreet, dignified prompt before overload: it's okay to leave now.
"How might we help people with social anxiety leave or reduce stimulation before overload through discreet AI cues that preserve autonomy and dignity?"
Research Insights
"At the moment of overload, a clear, empathetic message is worth far more than complex data or a polished UI."
Prototyping insightIn stressful social situations, people focus on their own nervous feelings rather than their surroundings. Crowds create escalating discomfort and loss of control, not sudden breakdown. Cognitive and attentional shifts precede full anxiety episodes.
Users notice subtle internal cues — tension, fatigue, increased vigilance — before social overload. These cues are frequently minimized: 'I should be fine' or 'it would be rude to leave now.'
Heart rate variability changes are reliable signals of emotional overload. But users raised concerns about misreadings, privacy, and feeling controlled by AI — revealing the opportunity: AI support must feel like a suggestion, not a directive.
Empathy Map
Jobs-to-be-Done
Jack attends a friend's birthday party with loud music and ongoing conversations. After socializing for a while, he begins to notice rising tension and difficulty focusing. Although increasingly overwhelmed, he stays — held back by social expectations.
Userflow Storyboard
A friend invites Jack to a party tonight. He dresses up and puts on his smartwatch before heading out.
The party starts well. Jack mingles and enjoys himself. His watch shows 12% pressure — all good.
As the night goes on, the crowd grows. Jack feels increasingly tired and overwhelmed. Pressure climbs to 70% — the watch vibrates gently.
The watch delivers a soft vibration pattern and a gentle prompt: 'Your signal says now might be a good time to leave.' A suggestion, not a command.
Jack calmly says goodbye to his friends — 'Sorry guys, I have to go.' No over-explanation needed. No guilt.
Jack gets home. The app logs the session data. He feels like he took care of himself — not that he ran away from something.
User Interface
Prototyping confirmed a core principle: at the moment of sensory overload, a clear and empathetic message is worth far more than complex data or a polished UI. SoftSignal's interface is deliberately minimal — a pressure reading, a short message, one action.
Pressure at 12%, green ring, 'All good' status. No action needed — user stays present and social.
All goodPressure at 70%, red ring, '⚠ Take a break' prompt. A gentle vibration pattern — no need to unlock a phone.
Take a breakUsers toggle Social Mode for context-aware pressure sensing, and Vibrate Mode for silent alerts — no notification sounds in public.
ConfigurableAI Mode analyzes real-time ambient conversations alongside biometric data to help users determine whether it might be a good time to leave — improving signal accuracy.
AI-assistedThe mobile app shows live pressure status and a history of past social events — each with duration and peak pressure. Users review data after the fact to build self-awareness, not stare at a screen in the moment.
Testing confirmed that people make more errors under pressure. At the moment of sensory overload, a clear, empathetic message is worth far more than complex data or a polished UI — driving the decision to keep the final interface radically minimal.
Analyzing the full interaction flow — from watch vibration patterns to notification messages to data collection — confirmed that privacy is paramount for this user group. AI prompts must feel like a gentle suggestion, never surveillance or control.
Testing revealed that emotional validation and tone matter far more than features like stress explanations or future recommendations. Users at high-pressure moments need to feel accepted — not analyzed.