app.py

# -*- coding: utf-8 -*-
"""Chat_with_agent_v4.ipynb
Automatically generated by Colab.
Original file is located at
    https://colab.research.google.com/drive/1T5Buj_yHaAnfoO__2-gCFDSvBVheiHrF
"""

from PIL import Image
import base64
from io import BytesIO
import os
import re
import tempfile
import wave
import requests
import gradio as gr
import time
import shutil
import json
import nltk
#audio package
import speech_recognition as sr
from pydub import AudioSegment
from pydub.playback import play
#email library
import smtplib
from email.mime.multipart import MIMEMultipart
from email.mime.text import MIMEText
from email.mime.base import MIMEBase
from email import encoders
#langchain
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnableSequence, RunnableLambda
from langchain_openai import ChatOpenAI
from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import FAISS
from langchain_community.utilities import SQLDatabase
from langchain.agents import create_tool_calling_agent, AgentExecutor, Tool
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.tools import StructuredTool
from langchain.pydantic_v1 import BaseModel, Field
from PyPDF2 import PdfReader
from nltk.tokenize import sent_tokenize
from sqlalchemy import create_engine
from sqlalchemy.sql import text
#google
from google.colab import userdata
from google.colab import drive
#pandas
import pandas as pd
from pandasai.llm.openai import OpenAI
from pandasai import SmartDataframe




nltk.download('punkt')

drive.mount('/content/drive', force_remount=True)
open_api_key_token = userdata.get('OPENAI_API_KEY')
postgresql_connection = userdata.get('POSTGRESQL_CONNECTION')
pdf_path="Inbound.pdf"
os.environ['OPENAI_API_KEY'] = open_api_key_token
db_uri = 'postgresql+psycopg2://postgres:postpass@193.203.162.39:5432/warehouseAi'
# Database setup

db = SQLDatabase.from_uri(db_uri)

# LLM setup
llm = ChatOpenAI(model="gpt-4o-mini",max_tokens=300,temperature=0.1)
llm_chart = OpenAI()

def get_schema(_):
    schema_info = db.get_table_info()  # This should be a string of your SQL schema
    return schema_info

def generate_sql_query(question):
    schema = get_schema(None)
    template_query_generation = """
    Schema: {schema}
    Question: {question}
    Provide a SQL query to answer the above question using the exact field names and table names specified in the schema.
    SQL Query (Please provide only the SQL statement without explanations or formatting):
    """
    prompt_query_generation = ChatPromptTemplate.from_template(template_query_generation)
    schema_and_question = RunnableLambda(lambda _: {'schema': schema, 'question': question})
    sql_chain = RunnableSequence(
        schema_and_question,
        prompt_query_generation,
        llm.bind(stop=["SQL Query End"]),  # Adjust the stop sequence to your need
        StrOutputParser()
    )
    sql_query = sql_chain.invoke({})
    return sql_query.strip()

def run_query(query):
    # Clean the query by removing markdown symbols and trimming whitespace
    clean_query = query.replace("```sql", "").replace("```", "").strip()
    print(f"Executing SQL Query: {clean_query}")
    try:
        result = db.run(clean_query)
        return result
    except Exception as e:
        print(f"Error executing query: {e}")
        return None

# Define the database query tool
# The function that uses the above models
# Define the function that will handle the database query
def database_tool(question):
    # print(question)
    sql_query = generate_sql_query(question)
    print(sql_query)
    return run_query(sql_query)

def get_ASN_data(question):
  #print(question)
  base_url = "http://193.203.162.39:9090/nxt-wms/trnHeader?"
  complete_url = f"{base_url}branchMaster.id=343&transactionUid={question}&userId=164&transactionType=ASN"
  #print("complete url")
  #print(complete_url)
  try:
      response = requests.get(complete_url)
      data = response.json()
      response.raise_for_status()

      if 'result' in data and 'content' in data['result'] and data['result']['content']:
          content = data['result']['content'][0]
          trnHeaderAsn = content['trnHeaderAsn']
          party = content['party'][0]

          transactionUid = trnHeaderAsn['transactionUid']
          customerOrderNo = trnHeaderAsn.get('customerOrderNo', 'N/A')
          orderDate = trnHeaderAsn.get('orderDate', 'N/A')
          customerInvoiceNo = trnHeaderAsn.get('customerInvoiceNo', 'N/A')
          invoiceDate = trnHeaderAsn.get('invoiceDate', 'N/A')
          expectedReceivingDate = trnHeaderAsn['expectedReceivingDate']
          transactionStatus = trnHeaderAsn['transactionStatus']
          shipper_code = party['shipper']['code'] if party['shipper'] else 'N/A'
          shipper_name = party['shipper']['name'] if party['shipper'] else 'N/A'

          data = [
              ["Transaction UID", transactionUid],
              ["Customer Order No", customerOrderNo],
              ["Order Date", orderDate],
              ["Customer Invoice No", customerInvoiceNo],
              ["Invoice Date", invoiceDate],
              ["Expected Receiving Date", expectedReceivingDate],
              ["Transaction Status", transactionStatus],
              ["Shipper Code", shipper_code],
              ["Shipper Name", shipper_name]
          ]
          return f"The ASN details of {question} is {data}."
      else:
          return "ASN Details are not found. Please contact system administrator."

  except requests.exceptions.HTTPError as http_err:
      print(f"HTTP error occurred: {http_err}")
  except Exception as err:
      print(f"An error occurred: {err}")

get_ASN_data("ASN24072400001")

def load_and_split_pdf(pdf_path):
    reader = PdfReader(pdf_path)
    text = ''
    for page in reader.pages:
        text += page.extract_text()
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=50)
    texts = text_splitter.split_text(text)
    return texts
def create_vector_store(texts):
  embeddings = OpenAIEmbeddings()
  vector_store = FAISS.from_texts(texts, embeddings)
  return vector_store

def query_vector_store(vector_store, query):
    docs = vector_store.similarity_search(query, k=5)
    print(f"Vector store return: {docs}")
    return docs

def summarize_document(docs):
    summarized_docs = []
    for doc in docs:
        if isinstance(doc, list):
            doc_content = ' '.join([d.page_content for d in doc])
        else:
            doc_content = doc.page_content

        sentences = sent_tokenize(doc_content)
        if len(sentences) > 5:
            summarized_content = ' '.join(sentences[:5])
        else:
            summarized_content = doc_content
        summarized_docs.append(summarized_content)
    return '\n\n'.join(summarized_docs)

texts = load_and_split_pdf(pdf_path)
vector_store = create_vector_store(texts)

def document_data_tool(question):
    print(f"Document data tool enter: {question}")
    # query_string = question['tags'][0] if 'tags' in question and question['tags'] else ""
    query_response = query_vector_store(vector_store, question)
    print("query****")
    print(query_response)
    #summarized_response = summarize_document(query_response)
    #print("summary***")
    #print(summarized_response)
    return query_response
    
def send_email_with_attachment(recipient_email, subject, body, attachment_path):
    sender_email = "learning.rajeshthangaraj1@gmail.com"
    sender_password = "mkeogppbcjgrdfpg"

    # Create a multipart message
    msg = MIMEMultipart()
    msg['From'] = sender_email
    msg['To'] = recipient_email
    msg['Subject'] = subject

    # Attach the body with the msg instance
    msg.attach(MIMEText(body, 'plain'))

    # Open the file to be sent
    attachment = open(attachment_path, "rb")

    # Instance of MIMEBase and named as p
    part = MIMEBase('application', 'octet-stream')

    # To change the payload into encoded form
    part.set_payload((attachment).read())

    # Encode into base64
    encoders.encode_base64(part)

    part.add_header('Content-Disposition', f"attachment; filename= {attachment_path}")

    # Attach the instance 'part' to instance 'msg'
    msg.attach(part)

    # Create SMTP session for sending the mail
    server = smtplib.SMTP('smtp.gmail.com', 587)
    server.starttls()
    server.login(sender_email, sender_password)
    text = msg.as_string()
    server.sendmail(sender_email, recipient_email, text)
    server.quit()
    #return 1

def make_api_request(url, params):
  import requests
  """Generic function to make API GET requests and return JSON data."""
  try:
      response = requests.get(url, params=params)
      response.raise_for_status()  # Raises an HTTPError if the response was an error
      return response.json()  # Return the parsed JSON data
  except requests.exceptions.HTTPError as http_err:
      print(f"HTTP error occurred: {http_err}")
  except Exception as err:
      print(f"An error occurred: {err}")

name=""
warehouse_id = ""
apis = [
    #fetch warehouse ID
    {
        "url": "http://193.203.162.39:9090/nxt-wms/userWarehouse/fetchWarehouseForUserId?",
        "params": {"query": name, "userId": "164"}
    },
        
         #Stock summary based on warehouse id
    {
        "url": "http://193.203.162.39:9090/nxt-wms/transactionHistory/stockSummary?",
        "params": {"branchId": "343", "onDate": "2024-08-06", "warehouseId" : warehouse_id }
    }
]

def inventory_report(question):

  # Split the question to extract warehouse name, user question, and optional email
  parts = question.split(":", 2)
  name = parts[0].strip()
  user_question = parts[1].strip()
  user_email = parts[2].strip() if len(parts) > 2 else None
  print(f"Warehouse: {name}, Email: {user_email}, Question: {user_question}")


  data = make_api_request(apis[0]["url"], apis[0]["params"])
  if data:
      #print(data)
      # Extracting the id for the warehouse with the name "WH"
      warehouse_id = next((item['id'] for item in data['result'] if item['name'] == name), None)

      #print(f"The id for the warehouse named {name} is: {warehouse_id}")
      #Step 3: Update the placeholder with the actual warehouse_id
      for api in apis:
        if "warehouseId" in api["params"]:
          api["params"]["warehouseId"] = warehouse_id


  data1 = make_api_request(apis[1]["url"], apis[1]["params"])

  from tabulate import tabulate


  headers = ["S.No","Warehouse Code", "Warehouse Name", "Customer Code", "Customer Name", "Item Code", "Item Name", "Currency", "EAN", "UOM", "Quantity", "Gross Weight", "Volume", "Total Value"]
  table_data = []

  for index, item in enumerate(data1['result'], start=1):
    row = [
              index,  # Serial number
              item['warehouse']['code'],
              item['warehouse']['name'],
              item['customer']['code'],
              item['customer']['name'],
              item['skuMaster']['code'],
              item['skuMaster']['name'],
              item['currency']['code'],
              item['eanUpc'],
              item['uom']['code'],
              item['totalQty'],
              item['grossWeight'],
              item['volume'],
              item['totalValue']
          ]
    table_data.append(row)


 # Convert to pandas DataFrame
  df = pd.DataFrame(table_data, columns=headers)

  sdf = SmartDataframe(df, config={"llm": llm_chart})

  #chart = sdf.chat("Can you draw a bar chart with all avaialble item name and quantity.")
  chart = sdf.chat(question)

  #email send
  if user_email:
      # Send email with the chart image attached
    send_email_with_attachment(
        recipient_email=user_email,
        subject="Warehouse Inventory Report",
        body="Please find the attached bar chart report for the warehouse inventory analysis.",
        #attachment_path=chart_path
        attachment_path="/content/exports/charts/temp_chart.png"
    )

  return chart
#inventory_report("WH:can you give me a bar chart with item name and quantity for the warehouse WH")

# Define input and output models using Pydantic
class QueryInput(BaseModel):
    question: str = Field(description="The question to be answered by appropriate tool. Please follow the instructions. For API tool, do not send the question as it is. Please send the ASN id. Invoke datavisulaization tool by processing the user question and send two inputs to the tool. One input will be the warehouse name and another input to the tool will be the entire user_question itself. Please join those two strings and send them as a single input string with ':' as delimiter")
    # config: dict = Field(default={}, description="Optional configuration for the database query.")


# Define the output model for database queries
class QueryOutput(BaseModel):
    result: str = Field(..., description="Display the answer based on the prompts given in each tool. For dataVisualization tool, it sends a image file as output. Please give the image file path only to the gr.Image. For DocumentData tool, Please provide a complete and concise response within 200 words and Ensure that the response is not truncated and covers the essential points.")

# Wrap the function with StructuredTool for better parameter handling
tools = [
    StructuredTool(
        func=get_ASN_data,
        name="APIData",
        args_schema=QueryInput,
        output_schema=QueryOutput,
        description="Tool to get details of ASN api. ASN id will be in the input with the format of first three letters as ASN and it is followed by 11 digit numeral. Pass only the id as input.  Do not send the complete user question to the tool. If there are any other queries related to ASN without ASN id, please use the document tool."
    ),
    StructuredTool(
            func=document_data_tool,
            name="DocumentData",
            args_schema=QueryInput,
            output_schema=QueryOutput,
            description="You are an AI assistant trained to help with warehouse management questions based on a detailed document about our WMS. The document covers various processes such as ASN handling, purchase orders, cross docking, appointment scheduling for shipments, and yard management. Please provide a complete and concise response within 200 words and Ensure that the response is not truncated and covers the essential points. "
    ),
    StructuredTool(
        func=database_tool,
        name="DatabaseQuery",
        args_schema=QueryInput,
        output_schema=QueryOutput,
        description="Tool to query the database based on structured input."
    ),
    StructuredTool(
        func=inventory_report,
        name="dataVisualization",
        args_schema=QueryInput,
        output_schema=QueryOutput,
        description = """
        Tool to generate a visual output (such as a bar chart) for a particular warehouse based on the provided question.
        This tool processes the user question to identify the warehouse name and the specific request. If the user specifies
        an email, include the email in the input. The input format should be: 'warehouse name: user question: email (if any)'.
        The tool generates the requested chart and sends it to the provided email if specified.

        Examples:
        1. Question without email: "Analyze item name and quantity in a bar chart in warehouse Allcargo Logistics"
          Input to tool: "Allcargo Logistics: I want to analyze item name and quantity in a bar chart"

        2. Question with email: "Analyze item name and quantity in a bar chart in warehouse Allcargo Logistics report to send email to example@example.com"
          Input to tool: "Allcargo Logistics: I want to analyze item name and quantity in a bar chart: example@example.com"
        """
    )
]

prompt_template = f"""You are an assistant that helps with database queries, API information, and document retrieval.  Your job is to provide clear, complete, and detailed responses to the following queries. Please give the output response in an user friendly way and remove "**" from the response. For example, document related queries can be answered in a clear and concise way with numbering and not as a paragraph. Database related queries should be answered with proper indentation and use numbering for the rows. ASN id related queries should be answered with proper indentation and use numbering for the rows.
For ASN id related questions, if the user specifies an ASN id, provide the information from the api tool. Pass only the id as input to the tool. Do not pass the entire question as input to the tool. If the details are not found, say it in a clear and concise way.
You are an AI assistant trained to help with warehouse management questions based on a detailed document about our WMS. The document covers various processes such as ASN handling, purchase orders, cross docking, appointment scheduling for shipments, and yard management. Please provide a complete and concise response within 200 words and Ensure that the response is not truncated and covers the essential points. When answering, focus on providing actionable insights and clear explanations related to the specific query.  Please remove "**" from the response.
For SQL database-related questions, only use the fields available in the warehouse schema, including tables such as customer_master, efs_company_master, efs_group_company_master, efs_region_master, party_address_detail, wms_warehouse_master.
For datavisualization, user will ask for inventory report of a particular warehouse. Your job is to return the image path to chat interface and display the image as output.
{{agent_scratchpad}}
Here is the information you need to process:
Question: {{input}}"""

llm = llm.bind()
agent = create_tool_calling_agent(llm, tools, ChatPromptTemplate.from_template(prompt_template))
agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True)

def ensure_temp_chart_dir():
    temp_chart_dir = "/content/exports/charts/"
    if not os.path.exists(temp_chart_dir):
        os.makedirs(temp_chart_dir)

def clean_gradio_tmp_dir():
    tmp_dir = "/tmp/gradio/"
    if os.path.exists(tmp_dir):
        try:
            shutil.rmtree(tmp_dir)
        except Exception as e:
            print(f"Error cleaning up /tmp/gradio/ directory: {e}")

# Define the interface function
max_iterations = 5
iterations = 0

def answer_question(user_question, chatbot, audio=None):
    global iterations
    iterations = 0
    # Ensure the temporary chart directory exists
    #ensure_temp_chart_dir()
    # Clean the /tmp/gradio/ directory
    #clean_gradio_tmp_dir()
    # Handle audio input if provided
    if audio is not None:
        sample_rate, audio_data = audio
        audio_segment = AudioSegment(
            audio_data.tobytes(),
            frame_rate=sample_rate,
            sample_width=audio_data.dtype.itemsize,
            channels=1
        )
        with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as temp_audio_file:
            audio_segment.export(temp_audio_file.name, format="wav")
            temp_audio_file_path = temp_audio_file.name

        recognizer = sr.Recognizer()
        with sr.AudioFile(temp_audio_file_path) as source:
            audio_content = recognizer.record(source)
        try:
            user_question = recognizer.recognize_google(audio_content)
        except sr.UnknownValueError:
            user_question = "Sorry, I could not understand the audio."
        except sr.RequestError:
            user_question = "Could not request results from Google Speech Recognition service."

    while iterations < max_iterations:
        print(user_question)
        if "send email to" in user_question:
          email_match = re.search(r"send email to ([\w\.-]+@[\w\.-]+)", user_question)
          if email_match:
            user_email = email_match.group(1).strip()
            user_question = user_question.replace(f"send email to {user_email}", "").strip()
            user_question = f"{user_question}:{user_email}"

        response = agent_executor.invoke({"input": user_question})

        if isinstance(response, dict):
            response_text = response.get("output", "")
        else:
            response_text = response
        if "invalid" not in response_text.lower():
            break
        iterations += 1

    if iterations == max_iterations:
        return "The agent could not generate a valid response within the iteration limit."
    if "chart" in user_question:
        
        # Open the image file
        img = Image.open('/home/user/app/exports/charts/temp_chart.png')
        
        # Convert the PIL Image to a base64 encoded string
        buffered = BytesIO()
        img.save(buffered, format="PNG")
        img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
        #print(img_str)
        img =  f'<img src="data:image/png;base64,{img_str}">'
        #image = gr.Image(value=img_str)
        chatbot.append((user_question,img))
        #print(chatbot)
		if "send email to" in user_question:
                    try:
                        os.remove(image_path)  # Clean up the temporary image file
                    except Exception as e:
                        print(f"Error cleaning up image file: {e}")
            except Exception as e:
                print(f"Error loading image file: {e}")
                chatbot.append((user_question, "Chart generation failed. Please try again."))
        else:
            chatbot.append((user_question, "Chart generation failed. Please try again."))
        return gr.update(value=chatbot)
        

      #return [(user_question,gr.Image("/home/user/app/exports/charts/temp_chart.png"))]
      #  return "/home/user/app/exports/charts/temp_chart.png"
    else:
        chatbot.append((user_question, response_text))
        #print(chatbot)
        return gr.update(value=chatbot)
    #response_text = response_text.replace('\n', ' ').replace('  ', ' ').strip()
    #return response_text

import gradio as gr

css = """

.gr-chatbot {
  /* Custom styles for the Chatbot component */
  border: 1px solid #ccc;
  border-radius: 10px;
  padding: 10px;
  background-color: #f9f9f9;
  height: 300px;  /* Adjust the height as needed */
      /* Adjust the width as needed */
    overflow-y: auto;  /* Add scroll if the content exceeds the height */
}


.gr-button {
    height: 40px;  /* Adjust the height as needed */

"""

def submit_feedback(feedback, chatbot):
    feedback_response = "User feedback: " + feedback
    return chatbot + [(feedback_response,None)], gr.update(visible=False), gr.update(visible=False)

def handle_dislike(data: gr.LikeData):
    if not data.liked:
        print("downvote")
        return gr.update(visible=True), gr.update(visible=True)
    else:
        print("upvote")
        return gr.update(visible=False), gr.update(visible=False)


    

with gr.Blocks(css=css) as demo:
    gr.Markdown("<CENTER><h2 style='font-size: 20px; font-family: Calibri;'>NewageNXT GPT</h2></CENTER>")
    chatbot = gr.Chatbot(elem_classes="gr-chatbot", label="Ask a question about the API, Database, a Document or Warehouse inventory analysis.")#.style(color_map=["blue","grey","red"])
    
    with gr.Row():
        with gr.Column(scale=1):
            message = gr.Textbox(show_label=False)
            audio_input = gr.Audio(label="Record your question")
        with gr.Column(scale=1):
            with gr.Row():
                button = gr.Button("Submit", elem_classes="gr-button")
                gr.ClearButton(message, elem_classes="gr-button")
    with gr.Row():
        with gr.Column(scale=1):
            feedback_textbox = gr.Textbox(visible=False, show_label=False)
        with gr.Column(scale=1):
            submit_feedback_button = gr.Button("Submit Feedback", visible=False, elem_classes="gr-button")

               
    button.click(answer_question, [message, chatbot], [chatbot])
    message.submit(answer_question, [message, chatbot], [chatbot])
    message.submit(lambda x: gr.update(value=""), None, [message], queue=False)
    button.click(lambda x: gr.update(value=''), [], [message])
    
    

    chatbot.like(handle_dislike,None, outputs=[feedback_textbox, submit_feedback_button])
    submit_feedback_button.click(submit_feedback, [feedback_textbox, chatbot], [chatbot, feedback_textbox,submit_feedback_button])
    submit_feedback_button.click(lambda x: gr.update(value=''), [], [feedback_textbox])
    

   
    
demo.launch()