app.py

from langchain.prompts import StringPromptTemplate
import re
import langchain
from qa_txt import conversation_chain
from key_extract import chain
from bs4 import BeautifulSoup
import requests
from data_process import *
from langchain.tools.base import StructuredTool
from langchain.agents import initialize_agent
from qa_txt import llm
import gradio as gr
from langchain.agents import (
    create_react_agent,
    AgentExecutor,
    tool,
)
from langchain import hub



prompt = hub.pull("hwchase17/react")

def faq(query: str) -> str:
    reponse = conversation_chain.invoke({"question": query, "chat_history": []})
    return reponse['answer']

qa_faq = StructuredTool.from_function(
    func = faq ,
    description="""
    Repondre à des questions general .

    Parameters :
    - query (string) : the same input as the user input no more no less and dont translate it even if it is in another language.

    Returns :
    - string : the output as returned from the function in french.
    """
)

def request_data(query: str) -> str:
    request = chain.invoke({"input": query})['text']
    mot_cle = nettoyer_string(request)
    mots = mot_cle.split()
    ui = mots[0]
    rg = chercher_data(ui)
    if len(rg[0]):
      reponse_final = format_reponse(rg)
      return reponse_final
    else:
      return "Désolé, il semble que nous n'ayons pas de données correspondant à votre demande pour le moment. Avez-vous une autre question ou avez-vous besoin d'aide sur quelque chose d'autre?"

fetch_data = StructuredTool.from_function(
    func=request_data,
    description="""
    Rechercher des données.

    Parameters :
    - query (string) : the same input as the user input no more no less and always must translate it in french if it isn't already. For example : "give me data about economy" the input is economy but need to be translated first in french and the inputed same for other languages.
    Returns :
    - string : the output as returned from the function in french.
    """,
)

tools_add = [
    qa_faq,
    fetch_data,
]

agent = create_react_agent(llm=llm, tools=tools_add, prompt=prompt)

agent_executor = AgentExecutor(
    agent=agent,
    tools=tools_add,
    verbose=True,
)

def data_gov_ma(message, history):
  try:
    response = agent_executor.invoke({"input": message})
    return response['output']
  except ValueError as e:
    return "Je suis désolé, je n'ai pas compris votre question. Pourriez-vous la reformuler s'il vous plaît ?"

gr.ChatInterface(data_gov_ma).launch()