Real-time FER + wellbeing pulse

A calmer way to read and reflect on your emotions.

AffectNet and CK+ models tuned for facial expressions, paired with a short wellbeing survey and an OpenAI-powered copilot to translate signals into caring suggestions.

Start a check-in Model & data

CLAHE lighting boost MTCNN or Haar fallback Local-first processing

How it flows

~15s demo

Capture or upload

Front-facing, clear lighting. We sanitize filenames and keep processing local.

Preprocess

CLAHE for contrast, MTCNN/Haar for detection, 48x48 grayscale crop normalized in-memory.

Predict & reflect

AffectNet/CK+ heads with flip-averaged logits; AI copilot writes a short, kind reflection.

Run it now

Model stack

AffectNet priority

Attempts AffectNet first, then CK+ JSON/weights, CK+-based H5, and finally legacy FER. Labels auto-align to output dims.

Wellbeing

Likert pulse

Four quick Likert questions; second is reverse-scored to catch stress drift. Score normalized to 0-20 and percent.

Privacy

Safe defaults

Sanitized filenames via Django storage, no raw keys in code, and minimal session footprint.

See privacy notes

Ready to see how you feel?

One photo + four answers → a concise, kind reflection.

Analyze emotion Wellness tips