Computer vision & assistive technology
SocialSync
Walk through the desktop app’s camera controls, four expression categories, and stabilized text feedback with inference off.
The application workflow
Connect camera input to stabilized expression feedback.
SocialSync is a team desktop prototype that connects camera input, local expression classification, and stabilized textual feedback. The interface below starts where the real application does: no camera, no session, and no inferred result.
- Role
- Full-stack development lead
- Application
- Electron + local Flask/TensorFlow service
- Integration work
- Camera-to-service workflow and temporal stabilization · 2026
Select a stage
The session begins only after the person enables the camera.
The original app asks for camera access here.
No frame is being captured or processed on this page.
What I built in the 2026 iteration
I connected the Electron interface to local inference and stabilized its output.
I connected the Electron camera interface to the local inference service, added session configuration and reset behavior, and carried the four-category output back into the desktop UI.
I added temporal stabilization around the changing class distribution, with transition handling and a recent-frame summary so the displayed feedback would not simply mirror each isolated frame.
Technical note: how stabilization works
The local service updates a session-specific, PID-inspired controller across the full four-class probability distribution. It combines the latest error, accumulated error, and rate of change, then clamps and normalizes the result.
A transition score can temporarily increase responsiveness. With no usable frame, the stored distribution decays toward an even state. These mechanics stabilize model output; they do not measure human emotion or establish a clinical outcome.