Inspiration This project was inspired by the need to better support people living with chronic illness. Today’s healthcare system is not designed for chronic conditions, patients often see multiple specialists, resulting in a fragmented and incomplete view of their health. In practice, some of the most valuable insights come from other patients with shared experiences. However, this information is scattered across outdated forums, Reddit threads, and siloed communities, making it difficult to access, trust, and act on.
What it does ChroniConnect is an AI-powered chronic illness copilot that helps patients: Describe their condition, symptoms, and severity
- Find similar patients using vector similarity search
- Receive AI-generated insights on what treatments worked, what didn’t, and key patterns
- Get actionable guidance to discuss with their healthcare provider
- Core idea:“Find what worked for patients like you.”
How we built it ChroniConnect is built as a full-stack AI system:
- Frontend: React (Vite) for a fast, interactive user experience
- Backend: FastAPI for handling queries and orchestration
- Embeddings: Sentence Transformers (all-MiniLM-L6-v2) to represent patient profiles
- Vector Search: FAISS for fast similarity matching across patient data
- LLM: Featherless AI (Qwen 2.5) for summarizing insights and generating recommendations
- Data Pipeline: Synthetic patient data generation + indexing pipeline
- All synthetic Data was generated by FeatherLess.ai API calls
Pipeline flow: User input → embedding → FAISS retrieval → LLM insight generation → personalized output What’s next for ChroniConnect
- Integrate GNN-based patient graph modeling to better capture relationships between patients beyond text similarity
- Build a scrollable, short-form content feed (TikTok/Reels-style) for intuitive discovery of patient experiences
- Launch a mobile iOS app for accessibility and real-time engagement Incorporate real-world patient data and validation layers to improve trust and accuracy
Built With
- faiss
- fastapi
- featherless
- react
Log in or sign up for Devpost to join the conversation.