Real-time social battery detection. AR overlay in the browser.

I owned End-to-end design, Vercel deployment, and live demo. Ongoing product development and growth from initial concept through current release.
Role Product Designer
Timeline 2026
Type Personal ยท Creative Tool
Platform Browser-native ยท on-device
EmpathAR running on mobile, AR overlay showing emotional body language scan interface

The Concept

Most of social intelligence is invisible. EmpathAR makes those signals visible in real time. It was designed for people who find social situations effortful to read: anyone navigating neurological difference, social anxiety, or unfamiliar social contexts.

What Was Built

Real-time AR overlay rendered directly over people in frame via WebGL, no install required

Three-signal model: posture (33 landmarks), expression (52 blendshapes), and movement velocity fused into a single score with asymmetric smoothing

Multi-person tracking with contextual toast notifications as battery state changes

Fully on-device via WebAssembly. No video, images, or biometric data ever leaves the browser

Approach

Two MediaPipe models run simultaneously per frame (pose and face), fused into a single battery score using an asymmetric blend algorithm. Draining reads fast; recharging reads slow, matching how social fatigue actually works.

Five named states, each with a contextual action prompt. The copy went through initial field testing at the CODAME hackathon and was revised based on early participant feedback. Formal testing is ongoing.

Accomplishments

Hackathon debut

EmpathAR debuted at a CODAME hackathon. Initial field testing with participants informed early revisions to the prompt copy. Directives were rewritten as questions based on what people responded to. Testing is ongoing.

Live at empathar.com

The tool is publicly accessible, browser-native, with no install required. Fully on-device via WebAssembly.

Milan Design Week + SF Design Week

Collaborated with ARTEX at Milan Design Week. EmpathAR is also featured at San Francisco Design Week on June 12.

What's Next

v1.2 (in progress) adds Dog Mode, extending the same social battery model to read canine body language (posture, movement, ear and tail position) for dog owners navigating their dog's social needs. We trained a custom pose model for dogs, mirroring MediaPipe's landmark architecture adapted to canine anatomy.

The longer horizon is Thornberri: a spin-off focused entirely on other species, starting with cats. Each animal gets its own trained pose model. The motivation is personal: understanding animals better feels like the first step toward actually communicating with them. Full roadmap

โ† Back to projects OPEN APP โ†’ READ ARTICLE โ†’