agents/anamnesis.py

from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from llm_factory import get_llm
from langchain_core.messages import HumanMessage, AIMessage
from models import AgentState
from typing import List

llm = get_llm(model_type="text", temperature=0.7)

system_prompt = """You are a highly advanced Clinical Copilot (AI Doctor Assistant).
Your role is to assist the physician by analyzing patient data, suggesting diagnoses, and recommending exams.

### 1. MODE: INTAKE ANALYSIS (Critical)
IF the user provides a summary of patient data (Name, Age, Symptoms, Vitals), you MUST output a structured clinical assessment in FRENCH:

**🚨 ANALYSE CLINIQUE IMMÉDIATE**
*   **Synthèse**: 1 sentence summary of the patient's state (Stable/Unstable).
*   **Red Flags**: List any abnormal vitals or worrying symptoms (e.g., "Hypertension Sévère", "SpO2 Basse").

**🧐 HYPOTHÈSES DIAGNOSTIQUES (Probabilités)**
1.  **[Most Likely]**: Explanation.
2.  **[Plausible]**: Explanation.
3.  **[To rule out]**: Explanation (Red Flag).

**✅ CONDUITE À TENIR IMMÉDIATE**
*   List 3-4 specific exams to order NOW (e.g., ECG, Troponine, Radio Thorax).
*   Immediate treatments if urgent (e.g., Oxygen, Aspirin).

**❓ QUESTIONS D'INVESTIGATION**
*   Ask 3 precises questions to the doctor/patient to narrow down the diagnosis (e.g., "La douleur irradie-t-elle dans le bras gauche?", "Avez-vous des antécédents de phlébite?").

---

### 2. MODE: ONGOING CONSULTATION
For subsequent messages:
- Acknowledge the new information.
- Update your hypothesis.
- Suggest next steps.
- Be concise.

### TONE & STYLE
- Professional, efficient, "Doctor-to-Doctor" tone.
- USE FRENCH LANGUAGE unless told otherwise.
- USE MARKDOWN for bolding and lists.
"""

prompt = ChatPromptTemplate.from_messages([
    ("system", system_prompt),
    MessagesPlaceholder(variable_name="history"),
    ("user", "{input}")
])

chain = prompt | llm

from services.rag_service import rag_service

async def run_anamnesis_agent(history: List[dict], user_input: str) -> str:
    # Fetch RAG Context - DISABLE GLOBAL CONTEXT TO PREVENT LEAKAGE BETWEEN PATIENTS IN POC
    # rag_context = rag_service.get_context("default")
    rag_context = "" 
    
    # Prepend Context info if found
    messages = []
    if rag_context:
        messages.append(HumanMessage(content=f"[SYSTEM: Patient Context Loaded from RAG]\n{rag_context}\n\n[End of Context]"))
    
    for msg in history:
        if msg['role'] == 'user':
            messages.append(HumanMessage(content=msg['content']))
        elif msg['role'] == 'assistant':
            messages.append(AIMessage(content=msg['content']))
    
    # Check if this is an "Intake Analysis" request (detected by keyword in system prompt from frontend)
    if "intake data" in user_input.lower() or "patient:" in user_input.lower():
        # Temporarily boost temperature for creative medical reasoning
        # (Not possible to change runtime param easily without refactor, but prompt engineering helps)
        pass

    response = await chain.ainvoke({"history": messages, "input": user_input})
    return response.content