SumiSense - On-Device Behavioral Health Assistant
On-device behavioral health assistant for private check-ins, trend detection, and safer clinician sharing.
Why I Built This
Privacy is foundational in recovery journeys, especially for people managing substance use disorders. During my SHARES Lab research, I repeatedly saw that fear of exposure or legal consequences can block the honesty needed for treatment. I built SumiSense (sum-sense, or breath-sense) to create a private path for reflection and early signal tracking without sending vulnerable text to the cloud. For me, this extends my broader work in privacy-preserving clinical decision support as I prepare to defend my master’s thesis.
What It Does
- Local reflection: Users write a 1-3 sentence check-in, and a local Qwen model extracts signals such as stress, sleep disruption, and cravings.
- Trend detection: Chronos is used on-device to detect instability drift across 14-day and 30-day windows instead of reacting to one difficult day.
- Safer sharing: A local anonymizer masks identifiers when users choose to share with clinicians, producing a research-safe summary.
Hackathon Outcome
SumiSense won 1st place at the ZETIC Melange On-Device AI Hackathon.
Links
- Demo video: Watch the demo
- GitHub repository: SumiSense