HEALTHCARE_IMPLEMENTATION_PLAN.md

🏥 Healthcare Chatbot – Implementation Flow (Dynamic & CPU-Optimized)

High-Level Runtime Flow

1. Message Intake: Socket.IO/API -> chat_messages -> chat_sessions
2. Context Resolution: context_service.py (Patient context + session memory)
3. Medical Intent Detection: intent_classifier.py (Symptom, Info, Safety, Casual, Non-medical)
4. Dynamic Risk Scoring: risk_engine.py (Score 0.0-1.0) -> Updates patient context
5. Dataset Routing: dataset_router.py (Weighted routing, e.g. SymCAT/MedQuAD)
6. Hybrid Retrieval: retrieval_engine.py (BM25 + FAISS + Cached Embeddings)
7. Safety & Compliance: safety_guard.py (No diagnosis, Prescriptions, Dosage; Add Disclaimers)
8. Response Composition: response_composer.py (Template vs LLM)
9. Delivery: Socket.IO -> storage -> feedback loop

Component Mapping

Existing Component	New Role
ai_engine.py	Medical Orchestrator Wrapper
FAISS	Primary Medical Retrieval
BM25	Clinical Term Matching
Unanswered Questions	Human-in-the-loop
Gemini	Last-resort Synthesis

Optimization

• CPU-Only: No heavy local LLMs (use APIs or efficient retrieval)
• Async: Non-blocking calls
• Caching: Embeddings and frequent responses

📊 Dataset & Knowledge Strategy

Dataset	Purpose	Notes
MedQuAD	Medical Q&A pairs	Covers diseases like flu, diabetes, cardiovascular; used for evidence-based responses
SymCAT	Symptom-to-Disease mapping	Helps chatbot compute risk score, suggest next questions, and verify symptom consistency
HealthTap	Professional health advice	Used to improve natural phrasing and context-aware responses
WHO / CDC	Guidelines & Escalation	Provides authoritative references for high-risk symptoms and emergency protocols
ICD-10 Codes	Disease Taxonomy	Standardized identifiers for internal mapping & consistent terminology

Next Steps

1. Implement medical_orchestrator.py (Core Brain)
2. Implement risk_engine.py with dynamic formulas
3. Add patient_context DB migration
4. Create healthcare-safe response templates
5. Build symptom stepper UI