felix Ai

Felix AI was inspired by the high-pressure environment of emergency rooms, where quick and accurate patient prioritization can make a life-saving difference. Traditional systems like the Emergency Severity Index (ESI) rely on manual judgment, which can vary under stress, so we set out to build an AI-powered solution that is faster, more consistent, and transparent. Felix AI collects patient symptoms and basic details, generates intelligent follow-up questions, and assigns an urgency score from 1 to 5 along with a clear explanation, while dynamically maintaining a live queue that ensures critical patients are prioritized first and providing concise pre-visit summaries for doctors. We built the system using Python with FastAPI as the backend and integrated the Anthropic Claude API (claude-sonnet-4-20250514) for reasoning, using in-memory storage for speed and a lightweight frontend with auto-refresh to simulate real-time updates. Along the way, we faced challenges such as enforcing clean JSON responses from the AI, handling API latency, and ensuring queue consistency, which we solved through structured prompt design, fallback mechanisms, and efficient sorting logic. We are proud to have built a system that not only prioritizes patients in real time but also explains its decisions, making it more trustworthy and practical for healthcare use. Through this project, we learned the importance of prompt engineering, system reliability, and simplicity in design, especially in time-sensitive applications. Moving forward, we aim to enhance Felix AI with live WebSocket updates, deeper integrations with healthcare platforms like Epic Systems, multilingual support, and advanced analytics to evolve it into a scalable, production-ready triage assistant.

Built With

• javascript
• mongodb
• nextjs
• postgresql
• python
• react