SumiSense - On-Device Behavioral Health Assistant

Privacy-first iOS assistant for daily check-ins, longitudinal trend detection, and safer clinician sharing.

Why I Built This

Privacy is foundational in recovery, especially for people managing substance use disorders. During my SHARES Lab research, I repeatedly saw how fear of exposure or legal consequences can make honest reflection harder. I built SumiSense to offer a private path for daily check-ins, early pattern awareness, and safer sharing without sending raw journal text to the cloud. The name draws on the Korean word for breath, and the project grew directly from my master’s work in privacy-preserving clinical decision support.

What It Does

  1. Local reflection: Users write a 1-3 sentence check-in. A local Qwen3-4B model surfaces signals such as stress, sleep disruption, cravings, isolation, and fatigue.
  2. Trend detection: Chronos compares 14-day and 30-day patterns so the app can surface gradual instability rather than overreacting to one difficult day.
  3. Safer sharing: A local anonymizer creates personal, clinician, or research-safe summaries while masking identifiers before export.

How I Built It

  • SwiftUI interface with an MVVM architecture and local persistence.
  • ZETIC MLange adapters for language, time-series, and anonymization models.
  • Deterministic fallback services for journal analysis, trends, and redaction when live models are unavailable.
  • Clear runtime status and a non-diagnostic boundary so users can see what is running locally.

Hackathon Outcome

SumiSense won 1st place at the ZETIC Melange On-Device AI Hackathon.

SumiSense is a pattern-awareness and safe-sharing tool. It does not diagnose, treat, or replace clinical care.