5 multimodal modules — Spotify, Social, Essay, Academic, Facial
ResNet stress detector at 85% accuracy
CatBoost academic-stress classifier at 90% accuracy
LLM essay sentiment reasoning at 82% accuracy
Five-module mental-health platform delivering a single multimodal wellbeing signal. Each module is a standalone Flask backend: (1) Spotify — audio-feature trends mapped to mood; (2) Social Media — YouTube/Reddit usage patterns; (3) Essay — LLM-driven sentiment reasoning over free-text (82% sentiment accuracy); (4) Academic — CatBoost classifier hitting 90% on academic stress; (5) Facial — ResNet stress detector at 85% on FER-style data. The Next.js frontend stitches everything together with Clerk AI auth + OAuth 2.0, MongoDB for state, and a unified dashboard that triangulates risk across signals rather than trusting any single modality.
