Event
Melange On-Device AI Hackathon
Team
Abhishek Singh Dhadwal
SumiSense is a privacy-first behavioral health companion designed to help users track emotional well-being without exposing sensitive personal information. Built around on-device AI, the platform transforms short daily journal entries into meaningful wellness insights while keeping raw notes entirely on the user’s device. By combining local language models, time-series forecasting, and privacy-preserving sharing tools, SumiSense enables users to monitor mental health patterns safely and securely.
What it solves
Behavioral health data is among the most sensitive forms of personal information, yet many existing wellness applications rely heavily on cloud infrastructure. This creates privacy concerns for users managing recovery, anxiety, stress, or substance-use challenges. SumiSense addresses this problem by allowing users to record honest reflections, identify wellness trends, and share only necessary insights without exposing raw journal entries or personal identifiers. The platform is designed around privacy regulations and the reality that trust is essential for meaningful self-reporting.
What it does
SumiSense converts short daily check-ins into private wellness signals that help users identify trends related to stress, sleep disruption, isolation, recovery, and emotional stability. The application provides local analysis, monitors changes over time, detects emerging patterns before they escalate, and generates privacy-safe summaries that can be shared with clinicians, therapists, or researchers. Users maintain complete control over what information leaves their device while still benefiting from AI-powered insights and long-term wellness tracking.
How it works
SumiSense combines multiple on-device AI capabilities powered through ZETIC Melange. Daily journal entries are analyzed locally using language models such as MedGemma and Qwen to generate wellness signals and trend tags. Time-series forecasting models monitor behavioral patterns across 14-day and 30-day windows to identify shifts in recovery, stress, and stability. When users choose to share information, an on-device anonymization pipeline automatically removes sensitive identifiers and generates privacy-safe summaries. By keeping inference, forecasting, and anonymization entirely on-device, SumiSense delivers meaningful behavioral health insights without relying on cloud-based processing.