Hea Helper

Inspiration

We live in a world of reactive healthcare—we only go to the doctor when we are already sick. But our bodies often whisper before they scream. We were inspired to create Hea Helper to listen to those whispers. The goal was to shift the paradigm from treatment to prevention by identifying weak signals of emerging health risks that might otherwise go unnoticed in the noise of daily life. We wanted to build a companion that helps users stay ahead of burnout, illness, and fatigue.

What it does

Hea Helper is a privacy-first wellness prototype that acts as an intelligent daily check-in.

1. Daily Logging: Users input unstructured notes about their day ("felt a bit off after lunch"), along with structured metrics like mood, sleep hours, and activity level.
2. AI Analysis: The app uses Google Gemini to analyze the current entry in the context of the user's recent history. It looks for patterns, anomalies, and contradictions (e.g., high activity but low mood).
3. Risk Detection: It assigns a risk level (Low to High) and detects specific "weak signals" (like "combined sleep debt with rising irritability").
4. Actionable Insights: Users receive a compassionate, non-medical explanation of the risk and a simple wellness recommendation to get back on track.

How we built it

We built Hea Helper with a focus on speed, aesthetics, and privacy.

• Frontend: We used React with Vite for a blazing fast local development experience.
• Styling: TailwindCSS allowed us to create a modern, "glassmorphism" aesthetic that feels calming and premium.
• Intelligence: The core logic is powered by the Google Gemini API. We engineered a system prompt that instructs the AI to act as a "Health Detector," teaching it to compare current data points against historical context to find subtle deviations.
• Visualization: We used Recharts to visualize wellness trends over time, making abstract data concrete for the user.

Challenges we ran into

• Prompt Engineering: Getting the AI to be helpful without hallucinating medical advice was tricky. We had to carefully tune the system prompt to focus on "wellness signals" rather than "diagnoses."
• Visualizing "Risk": We didn't want the app to feel alarmist. Balancing the color signs (Red/Orange/Green) so they felt informative but not panic-inducing required several design iterations.
• Data Context: Early versions analyzed only the current day, which missed the bigger picture. We had to implement a sliding window context (passing the last 3 days of history to the LLM) to truly detect trends rather than just events.

Accomplishments that we're proud of

• The "Vibe": We are really proud of the soothing UI. The ambient background blobs and glass panels make the app feel like a safe space for health data.
• Weak Signal Detection: Seeing Gemini correctly identify that a user's "grumpiness" was actually correlated with a drop in sleep two days prior was a clear "aha!" moment for the utility of this tool.
• Privacy-First Architecture: The app runs client-side logic and only sends anonymized prompt data to the API, respecting user privacy by design.

What we learned

• Context is King: AI is infinitely more useful when you give it history. Single-shot analysis is okay, but longitudinal analysis is where the magic happens.
• UX Matters for Health: If a health app feels clunky or clinical, people won't use it. Making it feel "soft" and approachable increases the likelihood of daily engagement.

What's next for Hea Helper

• Long-term Memory: Implementing a local database (like IndexedDB or PGlite) to store months of data for deeper trend analysis.
• Wearable Integration: Automatically pulling sleep and activity data from Apple Health or Google Fit to reduce manual entry friction.
• Personalized Models: Fine-tuning a smaller Gemini model specifically on the user's own baseline to reduce false positives.

Built With

• google-gemini-api
• lucide-react
• react
• recharts
• tailwindcss
• typescript
• vite