• 

  Chatbot + BrowserUse agent

• 

  MBTI style wellness quiz

• 

  Meal Plan

CarePilot

“Small signals. Big consequences. We know that you are busy. We will make them clear — before it’s too late.”

Inspiration

Over 40M people around the world ask ChatGPT daily for help with health concerns. However, ChatGPT usually treats each message in isolation. They don’t maintain a structured view of a user’s health across sleep, stress, digestion, and more—and they rarely help users take real-world action.

We wanted to build something that feels like a continuous companion:
it remembers your profile, reasons over it in chat, and can act—not just respond.

---

What It Does

CarePilot is a web app that combines a structured health profile, quick subhealth check, personalized chat, meal planning, and optional real-world execution via a browser agent.

Key Features

• Quick Check

  • Short Likert-style assessment with optional symptom tags
    
  • Generates a subhealth score
    
  • Saves to user profile (or browser for guests)

• Health Input

  • Body metrics + subhealth focus (1–5 scale)
    
  • Drives personalization for chat and meal planning
    

• Chat

  • Powered by Google Gemini (backend)
    
  • Profile-aware responses with conversation history
    
  • Optional RAG from a small knowledge corpus
    

• Recommendations (Actionable)

  • Structured steps generated by the assistant
    
  • Users can select steps and click Run selected
    
  • Executes via Browser Use Cloud (e.g., lookups, resources, grocery flows)

• Meal Plan

  • 7-day rolling plan starting from today
    
  • Dynamically updated from chat outputs
    

• Find Nearby

  • Google Maps API integration (when configured)
    
  • Search for groceries or care facilities
    
  • Clearly framed as “maps only / not medical advice”

turn information into plans and actions for users

How We Built It

Frontend

• Vite
  
• React + TypeScript
  
• React Router
  
• Tailwind-style utility classes + shared carepilot.css

Backend

• Node.js + Express (JavaScript)
  
• Google Gemini via @google/genai
  
• Browser Use Cloud integration
  
• Session-backed profiles
  
• REST APIs under /api/...

---

Challenges We Ran Into

Latency & Reliability

• Gemini + browser execution can be slow/fail
  
• Added:
  • Timeouts
    
  • Retries
    
  • Graceful fallbacks (e.g., mock planners)

Meaningful Context

• Raw memory ≠ useful personalization
  
• Solution:
  • Typed profile (scores, metrics)
    
  • Structured JSON outputs (meal updates, plans)

Chat → Execution Alignment

• Bridging assistant outputs to UI actions
  
• Built:
  • Step checklist
    
  • “Run selected” execution flow
    
  • Meal plan merging logic

---

Accomplishments

• Structured health profile + subhealth scoring system
  
• Context-aware Gemini (not generic chatbot replies)
  
• Actionable recommendations → executable browser steps
  
• Full working prototype:
  • Auth/session demo
    
  • Health input
    
  • Chat
    
  • Meal plan
    
  • Nearby search
    
  • Optional automation
    

---

What We Learned

• Structured data > unstructured memory for personalization
  
• Users want next steps, not just answers
  
• Combining UI + agent execution creates a stronger experience than chat alone
  

---

What’s Next

• Longitudinal tracking (trends over time)
  
• Deeper personalization from habits + goals
  
• Safer, more robust browser-agent workflows
  
• Reminders + lightweight progress dashboards tied to user profile
  

---

Summary

CarePilot turns AI wellness from passive advice into an interactive, personalized, and actionable system—bridging the gap between insight and real-world behavior.

Built With

• agent
• browseruse
• css
• gemini
• html
• javascript
• llm
• rag
• typescript