Overalhealth

Inspiration

We noticed that most health apps only track physical metrics, and finance apps only track spending — but nobody was connecting the two. We were inspired by the simple observation that what you spend money on directly reflects how healthy you live. A person buying fast food five times a week and skipping the gym tells a complete health story through their transactions alone. We wanted to build something that reads that story automatically.

What it does

OverallHealth is an AI-powered wellness dashboard that combines three data sources — daily transactions, fitness tracker data (Apple Watch), and manual health logs — to give users a complete picture of their health. It tags every transaction with a health impact (positive, neutral, or negative), tracks mood, sleep, energy and steps daily, finds correlations like "fast food spend drops your mood by 1.8 points the next day", and sends AI-generated weekly email reports with personalized recommendations powered by Claude.

How we built it

Frontend — React with Vite, custom dark UI with Syne and Instrument Sans fonts AI Agent — Claude API (claude-sonnet) with full transaction and health data baked into the system prompt Data — 30 synthetic transactions and 21 days of health logs with health impact tagging Fitness input — Apple Watch integration toggle that feeds steps, heart rate, sleep and calories into the AI Email reports — Claude generates the full report content, EmailJS SDK delivers it to the inbox Charts — Custom SVG bar charts and donut charts built from scratch

Challenges we ran into

Getting the Claude API to work directly from the browser required the anthropic-dangerous-direct-browser-access header which took time to figure out EmailJS kept returning 400 Bad Request errors because we were calling the REST API directly instead of using their official SDK Connecting Apple Watch data required simulating the HealthKit bridge since direct browser access to Apple Health is not possible without a native app Correlating transaction dates with health log dates to find meaningful patterns required careful data alignment Making the AI responses feel personal and data-specific rather than generic took several prompt iterations

Accomplishments that we're proud of

Built a fully working AI agent that has real context about your spending and health and can answer specific questions like "why was my energy low this week" The correlation engine successfully identifies real patterns between spending categories and health outcomes The AI-generated email report reads like it was written by a personal wellness coach, not a machine The entire app works as a single React file with no complex backend required Clean, professional dark UI that feels production-ready

What we learned

Transactions are a surprisingly powerful proxy for lifestyle and health habits Prompt engineering matters enormously — small changes to the system prompt completely changed the quality of AI health insights EmailJS SDK behaves very differently from calling their REST API directly — always use the official SDK Connecting financial data with biometric data creates insights that neither dataset could reveal alone Building with the Claude API directly in the browser is very fast for prototyping but needs a backend proxy for production

What's next for Overalhealth

Real bank integration using Plaid API to pull live transaction data automatically instead of synthetic data Native Apple Health and Google Fit sync via a React Native mobile app for real-time fitness tracker data Predictive health scoring — using 90 days of data to predict future health trends before they become problems Doctor report export — generate a PDF summary of health patterns to share with a physician Social challenges — let friends compare healthy spending habits and compete on wellness scores Subscription model — free tier for 30-day history, Pro tier for unlimited history, deeper AI analysis and priority email reports

Built With

• anthropic
• claude
• jss
• jsx
• llm
• react
• vite