Zzzync

About the project

zzzync was inspired by a very real hackathon situation: no sleep, too much caffeine, irregular food timing, stacked meetings, and a constant feeling of mental burnout. Most health apps show isolated stats, but they don’t answer the most important question fast enough: “Why am I tired right now?”

That became the core goal of this project.

I built zzzync as an iOS AI bio-sync coach that combines multiple personal signals into one causal explanation layer. Instead of giving generic wellness tips, it connects the dots across sleep debt, circadian drift, stimulant load, food timing, and schedule pressure, then gives concise actions that are actually usable in the moment.

At a high level, the project treats fatigue as a system problem:

[ \text{Fatigue pressure} \propto f(\text{sleep debt},\ \text{circadian misalignment},\ \text{stimulant load},\ \text{cognitive stress}) ]

How I built it

The app is built in Swift/SwiftUI with a local-first architecture. Data is pulled from:

HealthKit (sleep, HRV, resting heart rate)
Calendar/EventKit (timing and meeting load)
Food logs (including off-clock meal patterns and stimulant/sugar context)
Email stress signals and app-level forecast/protocol outputs

The AI layer is powered by Claude. I implemented both:

Structured analysis flows (for forecast/protocol style outputs)
Open chat-style Q&A for natural user questions

To make the AI demo trustworthy, I also added live response telemetry in chat (model used, token usage, and response ID), so it is visibly using Claude in real time.

Challenges I faced

Build config reliability: API key/model settings and ensuring they resolve correctly in iOS build environments
Claude integration hardening: fallback model handling and clearer failure paths
UX clarity: keeping responses short and visual instead of paragraph-heavy
Mobile interaction polish: keyboard/focus behavior in the AI tab affecting tab switching
Signal quality balance: grounding answers in real app data while keeping output concise