Update src/crew/crew_pool.py

src/crew/crew_pool.py

@@ -1,40 +1,27 @@
-from crewai import Crew, Process
-from langchain_core.tools import tool
 import json
-from pydantic import BaseModel, Field
-from typing import Dict, List, Any, Type
-from .agents import report_generator_agent, skills_extractor_agent, experience_extractor_agent, project_extractor_agent, education_extractor_agent, ProfileBuilderAgent, informations_personnelle_agent, reconversion_detector_agent
-from .tasks import generate_report_task, task_extract_skills, task_extract_experience, task_extract_projects, task_extract_education, task_build_profile, task_extract_informations, task_detect_reconversion
-from src.deep_learning_analyzer import MultiModelInterviewAnalyzer
-from src.rag_handler import RAGHandler 
-from langchain_core.tools import BaseTool
+from crewai import Crew, Process
+from .agents import report_generator_agent
+from .tasks import generate_report_task
+from typing import List
 
-@tool
-def interview_analyser(conversation_history: list, job_description_text: list) -> str:
+def run_interview_analysis(conversation_history: list, job_description_text: str, analyzer_model, rag_handler) -> str:
     """
-    Appelle cet outil à la toute fin d'un entretien d'embauche pour analyser
-    l'intégralité de la conversation et générer un rapport de feedback.
-    Ne l'utilise PAS pour répondre à une question normale, mais seulement pour conclure et analyser l'entretien.
+    Analyse l'intégralité de la conversation et génère un rapport de feedback.
+    Cette fonction est conçue pour être appelée en arrière-plan.
     """
-    # 1. Analyse DL de la conversation
-    analyzer = MultiModelInterviewAnalyzer()
-    structured_analysis = analyzer.run_full_analysis(conversation_history, job_description_text)
-
-    # 2. Enrichissement avec RAG
-    rag_handler = RAGHandler()
+    structured_analysis = analyzer_model.run_full_analysis(conversation_history, job_description_text)
     rag_feedback = []
-    # Extraire les intentions et sentiments pour trouver des conseils pertinents
     if structured_analysis.get("intent_analysis"):
         for intent in structured_analysis["intent_analysis"]:
-            # Exemple de requête basée sur l'intention
             query = f"Conseils pour un candidat qui cherche à {intent['labels'][0]}"
             rag_feedback.extend(rag_handler.get_relevant_feedback(query))
     
     if structured_analysis.get("sentiment_analysis"):
         for sentiment_group in structured_analysis["sentiment_analysis"]:
             for sentiment in sentiment_group:
-                if sentiment['label'] == 'stress' and sentiment['score'].item() > 0.6:
+                if sentiment['label'] == 'stress' and sentiment['score'] > 0.6:
                     rag_feedback.extend(rag_handler.get_relevant_feedback("gestion du stress en entretien"))
+
     unique_feedback = list(set(rag_feedback))
     interview_crew = Crew(
         agents=[report_generator_agent],
@@ -48,9 +35,23 @@ def interview_analyser(conversation_history: list, job_description_text: list) -
         'structured_analysis_data': json.dumps(structured_analysis, indent=2),
         'rag_contextual_feedback': "\n".join(unique_feedback)
     })
+    
     return final_report
 
-def analyse_cv(cv_content: str) -> json:
+def analyse_cv(cv_content: str) -> dict:
+    from .agents import (
+        informations_personnelle_agent, skills_extractor_agent, 
+        experience_extractor_agent, project_extractor_agent, 
+        education_extractor_agent, reconversion_detector_agent, 
+        ProfileBuilderAgent
+    )
+    from .tasks import (
+        task_extract_informations, task_extract_skills, 
+        task_extract_experience, task_extract_projects, 
+        task_extract_education, task_detect_reconversion, 
+        task_build_profile
+    )
+    
     crew = Crew(
         agents=[            
             informations_personnelle_agent,
@@ -59,7 +60,6 @@ def analyse_cv(cv_content: str) -> json:
             project_extractor_agent,
             education_extractor_agent,
             reconversion_detector_agent,
-
             ProfileBuilderAgent       
         ],
         tasks=[
@@ -76,6 +76,4 @@ def analyse_cv(cv_content: str) -> json:
         telemetry=False
     )
     result = crew.kickoff(inputs={"cv_content": cv_content})
-    return result
- 
-    
+    return json.loads(result)