Tool for measurement of digital biomarkers from video or audio of an individual’s behavior.

The Stanford Screenomics is an open-source Android app framework for capturing real-time digital trace data to support behavioral and health research.

Detect upcoming depressive and manic episodes in patients with bipolar disorder

Facial AU dynamics extraction (MediaPipe) + convolutional VAE phenotyping + speech fusion (OpenSMILE) for clinical interview analysis. R²=0.69 on synthetic data.

End-to-end behavioral prediction system using digital phenotyping. PyTorch Transformer (MAE 1.18) + Autoencoder anomaly detection. Docker-ready, FastAPI service.

Machine learning models for schizophrenia relapse prediction using digital phenotyping and passive smartphone sensing.

Real-time AI stress tracker using mouse & keystroke dynamics. Privacy-first, local-only, and 100% offline.

Deep learning-based early mental stress detection and 12-24h forecasting from passive behavioral phone data. Android + PyTorch. Academic research prototype.

OSRP (Open Sensing Research Platform) - Complete multi-modal mobile sensing for academic research. Built for AWS. Open source.

A biometric layer for telehealth. Turns the face and voice already present in a video visit into quantified, quality-aware measurements — computed in the browser, never recorded. Measures contrasts that cancel confounds (left vs. right, early vs. late) and abstains when the signal does not qualify. Nonclinical research prototype.

Termux-based Android event collector: offline-first SQLite outbox, idempotent authenticated batch upload to Home Assistant. Experimental.

An open-source prototype framework for smartphone-based digital phenotyping

Privacy-preserving Flutter SDK for behavioral signals — typing, motion, attention

Home Assistant app: authenticated ingestion API, event store, feature engineering, coverage metrics and a read-only MCP server for Android timeline data. Experimental.

Analysis of the DynAMoND study

Companion repository for the paper "Beyond Personalization: Cluster-Aware Shared Representation Learning for Wearable-Based Psychotic Relapse Detection"

NLP prediction of real-time suicide crisis states from daily text entries

Reproducible Python pipeline: psychiatric classification from wrist actigraphy. SHAP recovers delayed circadian phase in depression without clinical labels. Circadian timing separates disorders; activity volume separates patients from controls.