Health Twin -- Full Project Description
Inspiration
University life is unpredictable --- late nights, irregular meals, exam stress, social pressure, and constant cognitive load.
As shown in your slides (pages 2--9), students often don't realize they're crashing until it's too late. Small signals like sleep debt, stress spikes, inactivity, and poor nutrition build up silently.
Existing apps track just one thing at a time:\ Sleep apps. Step apps. Stress apps. Food apps.\ But no one connects the dots.
We asked:
"What if your health app didn't just track what you did...\ but actually understood you, anticipated risks, and protected you---automatically?"
That's how Health Twin was born.
A proactive, multi-agent digital twin that senses when your health is heading into the danger zone --- and steps in before real problems occur.
What It Does
Health Twin acts like a personal health radar.
It continuously listens to your day through sleep patterns, movement, stress signals, calendar load, and lifestyle events --- and builds a dynamic picture of your health.
It doesn't dump data on you.\ It gives you clarity, distilled into one simple idea:
"Here's what's happening → Here's what's coming → Here's what to do right now."
Based on your demo (page 17), it even talks to you:
- The agent calls you briefly.\
- You say what you ate or how you feel.\
- The digital twin updates itself automatically.
No tracking apps.\ No manual logging.\ Just a health system that notices and acts.
Key abilities:
- Detect rising stress before you feel it\
- Forecast sleep debt and recovery dips\
- Spot overload from upcoming exams or packed days\
- Deliver micro-interventions (rest, hydration, timing, planning)\
- Ping you with relevant advice or even make a quick call
If something looks dangerous, it escalates.\ If something looks manageable, it guides you.\ Always proactive. Always watching out for you.
How We Built It
We built Health Twin as a fully agentic, cloud-powered system designed to learn from real signals and proactively support students before problems arise.
Our approach combines a multi-agent AI backbone, a fast mobile-first interface, and a real-time data pipeline --- all designed to feel smooth, personal, and intelligent.
Agentic Intelligence with AWS Bedrock
At the core of Health Twin is a multi-agent architecture, powered by Amazon Bedrock.
Each agent specializes in one domain --- sleep, stress, fitness, burnout, or planning.
They collaborate through structured messages to evaluate the user's state, forecast risks, and generate personalized interventions.
This creates the feeling of a living, evolving digital twin rather than a static rules engine.
Frontend Built for Speed & Clarity
We built the interface using:
- React 18.3.1 for dynamic UI\
- TypeScript for reliability\
- Vite for instant builds\
- Tailwind CSS for clean, mobile-first design
Insightful Visualizations
To help users understand their health without feeling overwhelmed, we integrated:
- Recharts 2.15.4 for smooth, responsive visualizations\
- Mini-charts for quick metric snapshots\
- Trend lines, readiness indicators, and burnout risk cards
This makes the dashboard feel like a real-time health cockpit.
Backend & Data Layer
For fast, scalable data storage, we used:
- AWS Bedrock\
- n8n\
- A backend service layer to store:
- daily signals\
- agent recommendations\
- chat history\
- readiness calculations\
- user baselines
Everything is built with real flow in mind, not mockups.
Challenges We Ran Into
- Building a real digital twin required complex modeling of stress, recovery, and fatigue.\
- Multi-agent coordination introduced conflicts that needed arbitration.\
- Proactive intelligence needed to be helpful without being intrusive.\
- Medical safety required strong boundaries and escalation logic.\
- Frontend clarity demanded extreme simplicity in presenting health insights.\
- Voice-driven food logging (the agent calling feature) required real-time NLP + audio processing.
Accomplishments We're Proud Of
- Built a working digital twin that anticipates risks.\
- Created a multi-agent system that feels alive and adaptive.\
- Implemented real proactive interventions (WhatsApp, calls, notifications).\
- Designed a clean, intuitive UI that communicates health at a glance.\
- Demonstrated voice-driven nutrition logging through agent phone calls.\
- Developed a system that generalizes beyond students to professionals, parents, and athletes.
What We Learned
- Students need tools that think ahead, not more trackers.\
- Multi-modal signals reveal deep patterns that single apps miss.\
- Proactive AI (like instant calls) feels magical and supportive.\
- Safety is crucial --- the system must avoid diagnoses.\
- Multi-agent collaboration works brilliantly when roles are clear.
What's Next
- Advanced time-series models for burnout and stress forecasting\
- Wearable integrations (Apple Watch, Oura, Garmin, Whoop)\
- Better proactive planning for studying, sleep, and workload\
- Personal baseline learning for individual calibration\
- Doctor / therapist escalation\
- Expanding to EMEA + US\
- Evolving into a full Health OS for everyone
Team Members
Each founder brings unique technical, psychological, and entrepreneurial expertise.
Kuo-Yi Chao -- TUM PhD Candidate, AI Agents & Perception\ LinkedIn: https://www.linkedin.com/in/kuo-yi-chao/\
Luna-Marie Berszinski -- TUM IS, Programming Tutor\ LinkedIn: https://www.linkedin.com/in/lunaberszinski/\
Geart Ferhati -- TUM MGT Student, 3× Startup Founder\ LinkedIn: https://www.linkedin.com/in/geartferhati/\
Daniel Shamsi -- LMU CS & Psychology, Startup Founder\ LinkedIn: https://www.linkedin.com/in/daniel-shamsi/\
Built With
- agenticai
- ai
- love
Log in or sign up for Devpost to join the conversation.