Inspiration
The inspiration for Yuzu stems from the "Equitable Nutrition" gap. While high-income individuals use wearable rings and expensive concierge doctors to track biometrics, marginalized communities, particularly those on SNAP or WIC, are often left with generic advice that ignores their physiological reality. We wanted to create a "clinical-grade nutritionist in your pocket" that requires zero expensive hardware, only a basic webcam and a lab report.
What it does
Yuzu is an end-to-end health intelligence platform that correlates real-time biometrics with clinical lab data.
Bloodwork Intelligence: Users upload a PDF of their lab results. Yuzu identifies deficiencies (e.g., Vitamin D, Iron) and maps them to user goals like "Heart Health" or "Diabetes Management."
Real-Time Vitals: Using a high-performance C++ engine, Yuzu performs rPPG (Remote Photoplethysmography) to track Heart Rate and Breathing Rate via a webcam—no sensors required.
Prescriptive Nutrition: It recommends culturally diverse, budget-friendly meals that act as "stabilizers" for identified biomarkers, moving nutrition from "guessing" to "knowing."
How we built it
We engineered a sophisticated Hybrid Sidecar Architecture to ensure medical-grade performance on a web budget:
The Core: A native C++ SDK handles the high-frequency pixel analysis required for rPPG, running within Ubuntu (WSL2) to maintain system-level stability.
The Bridge: We built a custom Node.js relay to pipe biometric data from the Linux environment into our Next.js frontend via WebSockets.
The Brain: Gemini 2.5 serves as our multi-modal reasoning engine, parsing complex PDF lab reports and performing image recognition on meal photos to estimate nutrient density.
The Database: MongoDB Atlas stores documents that allow us to aggregate long-term biometric trends against caloric intake.
Challenges we ran into
Credential & Permission Friction: We faced significant "Credential Clash" issues while managing a multi-contributor GitHub environment, which we resolved by implementing Personal Access Tokens and streamlining our remote URLs.
Environment Bridging: Connecting a low-level Linux C++ engine to a high-level Windows Next.js dashboard presented networking hurdles (Localhost 403 errors), which we solved by creating a dedicated port-forwarding relay.
Data Volume: Managing a code-base that grew to 180k+ lines due to build artifacts required us to aggressively optimize our .gitignore and Git history to ensure the repository remained "accessible and well-organized" per the rubric.
Accomplishments that we're proud of
Zero-Hardware Biometrics: We successfully bypassed the "Wearable Tax" by getting medical-grade vitals from a standard laptop camera.
Cultural Equity: Our AI doesn't just suggest "Chicken and Broccoli"; it understands diverse flavor profiles and affordable ingredients, making personalized health accessible to everyone regardless of their economic background.
Full-Stack Integration: Seeing a PDF upload translate into a real-time "Biomarker Stability" gauge on a live dashboard was a major technical milestone for the team.
What we learned
Technical Resilience: We learned how to debug complex cross-platform environments (WSL to Windows) and the importance of clean Git hygiene in a fast-paced hackathon.
Human-Centric AI: We realized that AI is most powerful when it translates "scary" data (like low iron levels in a lab report) into "doable" actions (like a specific, tasty meal recommendation).
The Power of C++: We gained a deep appreciation for how native code can handle heavy mathematical lifting that browsers simply aren't built for.
What's next for Yuzu
Longitudinal Tracking: We plan to build out the MongoDB schema to show "Health Recovery Curves," proving to users that their dietary changes are physically improving their vitals over months.
Activity Sync: Sync Yuzu with apple health, Strava, or any other health tracking system.
Doctor-Ready Exports: A "One-Click Report" feature that packages biometric trends and meal adherence into a PDF that a user can take to their local clinic.
Direct SNAP Integration: A feature that allows users to scan their grocery receipts to see how their SNAP budget is being optimized for their specific biomarker deficiencies.
Finding Meals: Google maps API that can locate clean nutrition groceries or prepared meals in your area.
Log in or sign up for Devpost to join the conversation.