🩺 CareSketch – AI-Powered Personalized Care Plans

🌱 Inspiration

As someone who's witnessed family members struggle with caregiving — juggling schedules, medications, meals, and emotional stress — I wanted to build something meaningful.
CareSketch was born from the desire to make caregiving more structured, supportive, and emotionally considerate, especially for those without formal medical backgrounds.

💡 What it does

CareSketch takes natural language descriptions of a care scenario and generates a personalized daily care plan.
It includes:

• 📋 Medications with dosage and timing
  
• 🍽️ Suggested meals
  
• 🏃‍♀️ Light exercises and rest periods
  
• 🧠 Risk & red-flag detection
  
• 📆 An interactive 5-day timeline
  
• 💬 An empathetic emotional support chatbot
  
• 📄 PDF export of the full care plan

All generated with the help of a local LLaMA 3 model running via Ollama, so user privacy is preserved.

🛠️ How I built it

• Frontend: Streamlit for UI with gradient themes and chat functionality
• Backend:
  • Ollama (LLaMA 3) for local natural language reasoning
  • Custom prompt templates for structured JSON plan generation
  • Risk detection with simple rule-based parsing
  • Plotly for interactive timeline visualization
  • FPDF for exporting structured PDFs
• Extras:
  • Sidebar with input presets
  • Emotion-aware tone selection
  • Embedded chatbot using the same local model
  • Styled with custom CSS and config

🧗 Challenges I ran into

• Hosting Ollama and a large model on limited environments (Streamlit Cloud doesn’t support local LLMs)
• Managing prompt engineering to ensure consistent JSON outputs
• Preserving chat history and UI state in Streamlit without accidental refreshes
• Handling PDF export errors and formatting issues across devices

🏆 Accomplishments that I'm proud of

• Successfully integrated a local LLM pipeline with an interactive Streamlit app
• Created a friendly, emotional interface that helps caregivers feel heard
• Made the care plan visually engaging, structured, and downloadable
• Prioritized accessibility, offline capability, and privacy from the ground up

📚 What I learned

• Prompt tuning and JSON parsing with LLMs can be tricky — context and ordering matter a lot
• Streamlit is powerful but requires careful state management to avoid resets
• Ollama makes working with large models locally viable even for MVPs
• Emotional design can be embedded in functional tools through tone and UX

🚀 What's next for CareSketch

• 🔊 Add real-time voice input with Whisper for accessibility
• 🧠 Expand risk analysis with real medical guidelines
• 🕸️ Integrate EHR uploads or FHIR for medical professionals
• 🌐 Add multilingual support
• 💾 Package the app as a desktop tool or secure PWA for offline caregivers
• 🫂 Collaborate with elder care and hospice organizations for field testing

💖 Built with compassion and code at Hack the Vibe 2025
Let’s support caregivers — one care sketch at a time.

Built With

• fpdf
• llama3
• markdown
• ollama
• pandas
• plotly
• python
• streamlit
• subprocess