Update services/graph_service.py

services/graph_service.py

@@ -4,16 +4,15 @@ import json
 from typing import TypedDict, Annotated, Sequence, Dict, Any, List
 
 from langchain_openai import ChatOpenAI
+from langchain_core.runnables import Runnable
 from langchain_core.messages import BaseMessage, AIMessage, HumanMessage, SystemMessage
-from langchain.agents import create_openai_tools_agent, AgentExecutor
+from langchain.agents import create_openai_tools_agent
 from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
-# Nouveaux imports pour la gestion correcte du graphe
 from langgraph.graph import StateGraph, END
 from langgraph.prebuilt import ToolNode
 
 from tools.analysis_tools import trigger_interview_analysis
 
-
 class AgentState(TypedDict):
     messages: Annotated[Sequence[BaseMessage], lambda x, y: x + y]
     user_id: str
@@ -40,7 +39,7 @@ class GraphInterviewProcessor:
         self.skills_summary = self._extract_skills_summary()
         self.reconversion_info = self._extract_reconversion_info()
 
-        self.agent_executor = self._create_agent_executor()
+        self.agent = self._create_agent()
         self.graph = self._build_graph()
         logging.info("GraphInterviewProcessor initialisé avec succès.")
 
@@ -67,7 +66,7 @@ class GraphInterviewProcessor:
             return f"CANDIDAT EN RECONVERSION: {reconversion.get('analysis', '')}"
         return "Le candidat n'est pas identifié comme étant en reconversion."
 
-    def _create_agent_executor(self) -> AgentExecutor:
+    def _create_agent(self) -> Runnable:
         prompt = ChatPromptTemplate.from_messages([
             SystemMessage(content="{system_prompt_content}"),
             MessagesPlaceholder(variable_name="messages"),
@@ -75,10 +74,10 @@ class GraphInterviewProcessor:
         ])
         llm = ChatOpenAI(api_key=os.getenv("OPENAI_API_KEY"), model="gpt-4o-mini", temperature=0.7)
         tools = [trigger_interview_analysis]
-        agent = create_openai_tools_agent(llm, tools, prompt)
-        return AgentExecutor(agent=agent, tools=tools, verbose=True)
+        return create_openai_tools_agent(llm, tools, prompt)
 
     def _agent_node(self, state: AgentState):
+        """Prépare le prompt et appelle l'agent pour obtenir une décision (parler ou utiliser un outil)."""
         system_prompt_content = self.system_prompt_template.format(
             entreprise=self.job_offer.get('entreprise', 'notre entreprise'),
             poste=self.job_offer.get('poste', 'ce poste'),
@@ -91,14 +90,13 @@ class GraphInterviewProcessor:
             reconversion_analysis=self.reconversion_info
         )
         
-        response = self.agent_executor.invoke({
+        response = self.agent.invoke({
             "system_prompt_content": system_prompt_content,
             "messages": state["messages"],
             "user_id": state["user_id"],
             "job_offer_id": state["job_offer_id"],
-            "conversation_history": state["messages"]
         })
-        return {"messages": [response['output']]}
+        return {"messages": [response]}
 
     def _router(self, state: AgentState):
         """Décide du chemin à suivre après la réponse de l'agent."""
@@ -126,7 +124,7 @@ class GraphInterviewProcessor:
                 "end_turn": END
             }
         )
-        
+
         graph.add_edge("tools", "agent")
 
         return graph.compile()
@@ -141,24 +139,19 @@ class GraphInterviewProcessor:
             "messages": langchain_messages,
         }
         
-        logging.info(f"Invoking graph with initial state: {initial_state}")
         final_state = self.graph.invoke(initial_state)
-
-        logging.info(f"Graph finished. Final state: {final_state}")
         
         if not final_state or not final_state.get('messages'):
-            logging.error("Final state is empty or does not contain messages.")
+            logging.error("L'état final est vide ou ne contient pas de messages.")
             return {"response": "Erreur: Impossible de générer une réponse.", "status": "finished"}
 
         last_message = final_state['messages'][-1]
-        logging.info(f"Last message from graph: {last_message}")
         
         status = "finished" if hasattr(last_message, 'tool_calls') and last_message.tool_calls else "interviewing"
         
         response_content = last_message.content
-        logging.info(f"Final response content: '{response_content}', status: {status}")
         
         return {
             "response": response_content,
             "status": status
-        }
+        }