shisir2024/cureo_medi

Purpose and Runtime

MediAssist AI is a health-information assistant that accepts consultation input, applies safety-oriented triage and specialist analysis, and returns structured JSON. The Python backend is the primary runtime; static HTML/CSS/JavaScript pages provide patient, learning, and doctor-facing interfaces. Deployment is indicated by Procfile; local startup is supported by run.sh.

Paths: backend/main.py, Procfile, frontend/consultation.html

API Entry and Session Boundaries

The backend application is initialized in backend/main.py and exposes domain routers for agent execution, authentication, consultations, and doctor workflows. Route modules form the HTTP boundary between frontend clients and application services. WebSocket state/connection handling is separated for real-time interactions.

Paths: backend/main.py, backend/routers/consult_routes.py, backend/ws_manager.py

Authentication and Persistence

Authentication is isolated from consultation logic, while database access is centralized behind a dedicated module. This creates a boundary for user identity, protected doctor/patient workflows, and persisted consultation-related records without coupling agents directly to storage.

Paths: backend/auth.py, backend/database.py, backend/routers/auth_routes.py

Orchestration Pipeline

backend/orchestrator.py is the central control-flow layer. It coordinates specialized agents rather than exposing their implementation details to routes. The expected flow starts with triage, proceeds to symptom analysis, conditionally invokes diagnosis/treatment/referral and visualization/scoring agents, then compiles a final response.

Paths: backend/orchestrator.py, backend/agents/triage.py, backend/agents/report.py

Safety and Emergency Handling

Triage is the first decision stage and governs safety-sensitive branching. Emergency cases are escalated through symptom analysis and visualization-oriented outputs, with an emergency fallback path rather than routine diagnostic/treatment progression. Global safety behavior and response framing are likely shared through the system prompt and agent prompts.

Paths: backend/agents/triage.py, backend/agents/symptom_analyzer.py, backend/system_prompt.py

Specialist Agent Layer

Each clinical-information concern is encapsulated as an agent: diagnosis, treatment, referral, vitals, symptom web, health score, heatmap, radar, and comparison. This supports parallel or selectively invoked analysis stages under orchestrator control. The report agent is the aggregation boundary that converts heterogeneous agent results into the final structured JSON contract.

Paths: backend/agents/diagnosis.py, backend/agents/treatment.py, backend/agents/health_score.py

LLM and Client Integration

LLM provider access is centralized in backend/llm.py, keeping model/API-key integration outside individual route handlers and likely reusable across agents. The frontend uses standalone pages and browser JavaScript; consultation and doctor portal pages are distinct clients of backend APIs.

Paths: backend/llm.py, frontend/consultation.js, frontend/doctor.js