agent.py

import os
import sys
from typing import List, Union

from langchain.prompts import ChatPromptTemplate
from langchain.prompts.chat import MessagesPlaceholder
from langchain.schema import BaseMessage, AIMessage, HumanMessage, SystemMessage
from langchain.agents import AgentExecutor, create_structured_chat_agent
from langchain_google_genai import ChatGoogleGenerativeAI

from config.settings import settings
from services.logger import app_logger
from tools import BioPortalLookupTool, UMLSLookupTool, QuantumTreatmentOptimizerTool

# -----------------------------------------------------------------------------
# 1. Initialize the Gemini LLM
# -----------------------------------------------------------------------------
def _init_llm() -> ChatGoogleGenerativeAI:
    """
    Initialize the Google Gemini LLM with the configured API key.
    Raises ValueError if no key is found or initialization fails.
    """
    api_key = settings.GEMINI_API_KEY or os.getenv("GOOGLE_API_KEY")
    if not api_key:
        err = "Gemini API key not found: set GEMINI_API_KEY in settings or GOOGLE_API_KEY in env"
        app_logger.error(err)
        raise ValueError(err)

    try:
        llm = ChatGoogleGenerativeAI(
            model="gemini-1.5-pro-latest",
            temperature=0.2,
            google_api_key=api_key,
            convert_system_message_to_human=True,
        )
        app_logger.info(f"Gemini LLM initialized ({llm.model})")
        return llm
    except Exception as e:
        err = f"Failed to initialize Gemini LLM: {e}"
        app_logger.error(err, exc_info=True)
        raise ValueError(err)

# -----------------------------------------------------------------------------
# 2. Build the structured chat prompt
# -----------------------------------------------------------------------------
def _build_prompt_template(tool_names: List[str], tools) -> ChatPromptTemplate:
    """
    Construct a ChatPromptTemplate that includes:
    - a system instruction block,
    - a placeholder for chat_history (List[BaseMessage]),
    - the current human input,
    - a placeholder for agent_scratchpad (List[BaseMessage]) to manage tool calls.
    """
    system_text = """
You are Quantum Health Navigator, an AI assistant for healthcare professionals.

• Disclaim: you are an AI, not a substitute for clinical judgment.
• Patient context is provided as: {patient_context}
• Available tools: {tool_names}
{tools}

To call a tool, reply *only* with a JSON block:
{"action": "<tool_name>", "action_input": <input>}

After receiving the tool’s output, craft a full answer for the user, citing any tools used.
""".strip()

    return ChatPromptTemplate.from_messages([
        ("system", system_text),
        MessagesPlaceholder(variable_name="chat_history"),
        ("human", "{input}"),
        MessagesPlaceholder(variable_name="agent_scratchpad"),
    ])

# -----------------------------------------------------------------------------
# 3. Lazily build and return the AgentExecutor singleton
# -----------------------------------------------------------------------------
def get_agent_executor() -> AgentExecutor:
    """
    Returns a singleton AgentExecutor, creating it on first call.
    Sets up LLM, tools, prompt, and executor params.
    """
    global _agent_executor_instance
    if "_agent_executor_instance" not in globals():
        # 3.1 Initialize LLM
        llm = _init_llm()

        # 3.2 Prepare tools
        tools_list = [
            UMLSLookupTool(),
            BioPortalLookupTool(),
            QuantumTreatmentOptimizerTool(),
        ]
        app_logger.info(f"Loaded tools: {[t.name for t in tools_list]}")

        # 3.3 Build prompt
        prompt = _build_prompt_template(
            tool_names=[t.name for t in tools_list],
            tools=tools_list
        )
        app_logger.info("Prompt template built")

        # 3.4 Create the structured agent
        agent = create_structured_chat_agent(
            llm=llm,
            tools=tools_list,
            prompt=prompt
        )
        app_logger.info("Structured chat agent created")

        # 3.5 Create the executor
        executor = AgentExecutor(
            agent=agent,
            tools=tools_list,
            verbose=True,
            handle_parsing_errors=True,
            max_iterations=10,
            early_stopping_method="generate",
        )
        app_logger.info("AgentExecutor initialized")
        _agent_executor_instance = executor

    return _agent_executor_instance

# -----------------------------------------------------------------------------
# 4. Optional REPL for local testing
# -----------------------------------------------------------------------------
if __name__ == "__main__":
    try:
        executor = get_agent_executor()
    except Exception as e:
        print(f"❌ Initialization failed: {e}")
        sys.exit(1)

    # Sample patient context for testing
    patient_context = (
        "Age: 58; Gender: Female; Chief Complaint: Blurry vision & fatigue; "
        "History: Prediabetes, mild dyslipidemia; Medications: None."
    )
    chat_history: List[Union[SystemMessage, HumanMessage, AIMessage]] = []

    print("🚀 Quantum Health Navigator Console (type 'exit' to quit)")
    while True:
        user_input = input("👤 You: ").strip()
        if user_input.lower() in {"exit", "quit"}:
            print("👋 Goodbye!")
            break
        if not user_input:
            continue

        try:
            result = executor.invoke({
                "input": user_input,
                "chat_history": chat_history,
                "patient_context": patient_context
            })
            reply = result.get("output", "")
            print(f"🤖 Agent: {reply}\n")

            # Update history
            chat_history.append(HumanMessage(content=user_input))
            chat_history.append(AIMessage(content=reply))
            # Trim to last 20 messages
            if len(chat_history) > 20:
                chat_history = chat_history[-20:]
        except Exception as err:
            print(f"⚠️ Inference error: {err}")
            app_logger.error("Runtime error in REPL", exc_info=True)