interview_agents_api / src /interview_simulator /entretient_version_prod.py
QuentinL52's picture
Update src/interview_simulator/entretient_version_prod.py
28ace03 verified
raw
history blame
3.85 kB
import os
import sys
import json
from typing import Dict, List, Any, Annotated
from typing_extensions import TypedDict
from langchain_core.messages import AIMessage, SystemMessage, HumanMessage, ToolMessage
from langchain_groq import ChatGroq
from langgraph.graph import StateGraph, START, END
from langgraph.graph.message import add_messages
from langgraph.prebuilt import ToolNode
from langchain_openai import ChatOpenAI
from src.config import read_system_prompt, format_cv
from src.crew.crew_pool import interview_analyser
class State(TypedDict):
messages: Annotated[list, add_messages]
class InterviewProcessor:
def __init__(self, cv_document: Dict[str, Any], job_offer: Dict[str, Any], conversation_history: List[Dict[str, Any]]):
if not cv_document or 'candidat' not in cv_document:
raise ValueError("Document CV invalide fourni.")
if not job_offer:
raise ValueError("Données de l'offre d'emploi non fournies.")
self.job_offer = job_offer
self.cv_data = cv_document['candidat']
self.conversation_history = conversation_history
self.tools = [interview_analyser]
self.llm = self._get_llm()
self.llm_with_tools = self.llm.bind_tools(self.tools)
self.system_prompt_template = self._load_prompt_template()
self.graph = self._build_graph()
def _get_llm(self) -> ChatOpenAI:
openai_api_key = os.getenv("OPENAI_API_KEY")
return ChatOpenAI(
temperature=0.6,
model_name="gpt-4o-mini",
api_key=openai_api_key
)
def _load_prompt_template(self) -> str:
return read_system_prompt('prompts/rag_prompt_old.txt')
def _chatbot_node(self, state: State) -> dict:
if state["messages"] and isinstance(state["messages"][-1], ToolMessage):
tool_message = state["messages"][-1]
return {"messages": [AIMessage(content=tool_message.content)]}
messages = state["messages"]
formatted_cv_str = format_cv(self.cv_data)
mission = self.job_offer.get('mission', 'Non spécifiée')
profil_recherche = self.job_offer.get('profil_recherche', 'Non spécifié')
competences = self.job_offer.get('competences', 'Non spécifiées')
pole = self.job_offer.get('pole', 'Non spécifié')
system_prompt = self.system_prompt_template.format(
entreprise=self.job_offer.get('entreprise', 'notre entreprise'),
poste=self.job_offer.get('poste', 'ce poste'),
mission=mission,
profil_recherche=profil_recherche,
competences=competences,
pole=pole,
cv=formatted_cv_str
)
llm_messages = [SystemMessage(content=system_prompt)] + messages
response = self.llm_with_tools.invoke(llm_messages)
return {"messages": [response]}
def _route_after_chatbot(self, state: State) -> str:
last_message = state["messages"][-1]
if last_message.tool_calls:
return "call_tool"
return END
def _build_graph(self) -> any:
graph_builder = StateGraph(State)
graph_builder.add_node("chatbot", self._chatbot_node)
graph_builder.add_node("call_tool", ToolNode(self.tools))
graph_builder.add_edge(START, "chatbot")
graph_builder.add_conditional_edges(
"chatbot",
self._route_after_chatbot,
{
"call_tool": "call_tool",
END: END
}
)
graph_builder.add_edge("call_tool", "chatbot")
return graph_builder.compile()
def run(self, messages: List[Dict[str, Any]]) -> Dict[str, Any]:
initial_state = self.conversation_history + messages
return self.graph.invoke({"messages": initial_state})