Missesqueensy/DarijaBridge

Darija in, structured triage out — an AI pre-screening assistant for the language 30 million people speak and almost no AI understands. # 🇲🇦 DarijaBridge: Medical Intake Assistant for Moroccan Patients **DarijaBridge** is an intelligent, secure, and empathetic medical intake chatbot built specifically for Moroccan patients. Powered by a multi-agent orchestrated graph (**LangGraph**) and running entirely on **Groq**, it bridges the language gap in healthcare by interacting with patients in pure Moroccan Darija (both Arabic and Arabizi scripts) to collect clinical history before structuring it into professional medical summaries. --- ## 🚀 Key Features * **100% Moroccan Darija Support:** Engineered using custom localized safeguards to stay strictly within authentic Moroccan vocabulary, completely blocking neighboring or Middle Eastern dialects. * **Multi-Agent Orchestration:** Driven by **LangGraph** to manage real-time conversation state tracking, context-aware routing, and safe termination structures. * **Strict Clinical Safety Boundaries:** Programmed to never suggest a medical diagnosis, never mention medications, and focus solely on gathering physiological complaints to support triage. * **Automatic Triage Summary:** Once adequate symptoms (location, progression, duration) are verified, it automatically generates a structured clinical note in French/English for the healthcare professional. --- ## 🏗️ Architecture & Graph Logic The system is organized as a dynamic workflow state machine: 1. **`chat_node`**: Manages the patient dialogue loop using low-latency streaming endpoints via `llama-3.3-70b-versatile`. 2. **`router_node`**: A deterministic and evaluation-based router that checks the current dialog turn and verifies whether the chief complaint, location, and timeline have been gathered. 3. **`summarize_node`**: Compiles the entire interaction history into a structured intake profile featuring complaint translations, duration mappings, and basic triage urgency levels. --- ## 🛠️ Project Structure ```text ├── agent.py # Core LangGraph implementation & model no …