app.py

import streamlit as st
from tempfile import NamedTemporaryFile

import pprint
import os
from dotenv import load_dotenv, find_dotenv
import os
from langchain_openai import ChatOpenAI
from langchain_community.document_loaders import PyPDFLoader
from langchain.document_loaders.csv_loader import CSVLoader
from langchain.document_loaders import WebBaseLoader

import pandas as pd
import numpy as np
import pprint

defaultGoogleURL = "https://www.google.com/search?q=google+earnings"
OPEN_ROUTER_KEY = st.secrets["OPEN_ROUTER_KEY"]
OPEN_ROUTER_MODEL = "meta-llama/llama-3.1-70b-instruct:free"

def pretty_print_columns(text):
    """
    Beautifies the provided CSV column description text.
    
    Args:
        text (str): The input string containing the column descriptions.
    
    Returns:
        str: The beautified string with neatly formatted column descriptions.
    """
    return "  ".join([line.strip() for line in text.splitlines() if line.strip()])

isPswdValid = True
try:
    pswdVal = st.experimental_get_query_params()['pwd'][0]
    if pswdVal==st.secrets["PSWD"]:
        isPswdValid = True
except:
    pass

if not isPswdValid:
    st.write("Invalid Password")
else:
    radioButtonList = ["E-commerce CSV (https://www.kaggle.com/datasets/mervemenekse/ecommerce-dataset)",
    "Upload my own CSV",
    "Upload my own PDF",
    f"URL Chat with Google's Latest Earnings ({defaultGoogleURL})",
    "Enter my own URL"]

    # Add some designs to the radio buttons
    st.markdown("""
    <style>
    .stRadio {
      padding: 10px;
      border-radius: 5px;
      background-color: #f5f5f5;
    }

    .stRadio input[type="radio"] {
      position: absolute;
      opacity: 0;
      cursor: pointer;
    }

    .stRadio label {
      display: flex;
      justify-content: center;
      align-items: center;
      cursor: pointer;
      font-size: 16px;
      color: #333;
    }

    .stRadio label:hover {
      color: #000;
    }

    .stRadio.st-selected input[type="radio"] ~ label {
      color: #000;
      background-color: #d9d9d9;
    }
    </style>
    """, unsafe_allow_html=True)

    genre = st.radio(
        "Tired of reading your files? Chat with it using AI! Choose dataset to finetune", radioButtonList, index=0
    )

    # Initialize language model
    load_dotenv(find_dotenv()) # read local .env file
    llm = ChatOpenAI(model=OPEN_ROUTER_MODEL, temperature=0.1, openai_api_key=OPEN_ROUTER_KEY, openai_api_base="https://openrouter.ai/api/v1")

    pdfCSVURLText = ""
    if genre==radioButtonList[0]:
        pdfCSVURLText = "CSV"
        exampleQuestion = "Question1: What was the most sold item? Question2: What was the most common payment?"
        loader = CSVLoader(file_path='EcommerceDataset.csv')
        csv_data = loader.load()
        # st.write('You selected comedy.')
        # else:
        # st.write(f'''Password streamlit app: {st.secrets["PSWD"]}''')
    elif genre==radioButtonList[1]:
        pdfCSVURLText = "CSV"
        exampleQuestion = "What are the data columns?"
    elif genre==radioButtonList[2]:
        pdfCSVURLText = "PDF"
        exampleQuestion = "Can you summarize the contents?"
    elif genre==radioButtonList[3]:
        pdfCSVURLText = "URL"
        exampleQuestion = "What is Google's latest earnings?"
    elif genre==radioButtonList[4]:
        pdfCSVURLText = "URL"
        exampleQuestion = "Can you summarize the contents?"

    isCustomURL = genre==radioButtonList[4]
    urlInput = st.text_input('Enter your own URL', '', placeholder=f"Type your URL here (e.g. {defaultGoogleURL})", disabled=not isCustomURL)

    isCustomPDF = genre==radioButtonList[1] or genre==radioButtonList[2]
    uploaded_file = st.file_uploader(f"Upload your own {pdfCSVURLText} here", type=pdfCSVURLText.lower(), disabled=not isCustomPDF)
    uploadedFilename = ""
    if uploaded_file is not None:
        with NamedTemporaryFile(dir='.', suffix=f'.{pdfCSVURLText.lower()}') as f:
            f.write(uploaded_file.getbuffer())
            uploadedFilename = f.name
            if genre==radioButtonList[1]: # Custom CSV Upload
                loader = CSVLoader(file_path=uploadedFilename)
                csv_data = loader.load()
            elif genre==radioButtonList[2]: # Custom PDF Upload
                loader = PyPDFLoader(uploadedFilename)
                pdf_pages = loader.load_and_split()

    enableChatBox = False
    if genre==radioButtonList[0]: # E-commerce CSV
        enableChatBox = True
    elif genre==radioButtonList[1]: # Custom CSV Upload
        enableChatBox = uploadedFilename[-4:]==".csv"
    elif genre==radioButtonList[2]: # Custom PDF Upload
        enableChatBox = uploadedFilename[-4:]==".pdf"
    elif genre==radioButtonList[3]: # Google Alphabet URL Earnings Report
        enableChatBox = True
    elif genre==radioButtonList[4]: # Custom URL
        enableChatBox = True

    chatTextStr = st.text_input(f'Ask me anything about this {pdfCSVURLText}', '', placeholder=f"Type here (e.g. {exampleQuestion})", disabled=not enableChatBox)
    chatWithPDFButton = "CLICK HERE TO START CHATTING"
    if st.button(chatWithPDFButton, disabled=not enableChatBox and not chatTextStr): #  Button Cliked
        if genre==radioButtonList[0]: # E-commerce CSV
            # Initializing the agent
            answer = llm.predict(f'''
            I have CSV file contents below:

            {str(csv_data)}

            {chatTextStr}
            ''')
            st.write(answer)

        elif genre==radioButtonList[1]: # Custom CSV Upload
            # Initializing the agent
            answer = llm.predict(f'''
            I have CSV file contents below:

            {str(csv_data)}

            {chatTextStr}
            ''')
            st.write(answer)

        elif genre==radioButtonList[2]: # Custom PDF Upload
            pdf_answer = llm.predict(f'''
            I have PDF file contents below:

            {str(pdf_pages)}

            {chatTextStr}
            ''')
            st.write(pdf_answer)
        elif genre==radioButtonList[3]: # Google Alphabet URL Earnings Report
            loader = WebBaseLoader(defaultGoogleURL)
            web_data = loader.load()
            answer = llm.predict(f'''
            I have website contents below:

            {str(web_data)}

            {chatTextStr}
            ''')

            st.write(answer)
        elif genre==radioButtonList[4]: # Custom URL
            loader = WebBaseLoader(urlInput)
            web_data = loader.load()
            answer = llm.predict(f'''
            I have website contents below:

            {str(web_data)}

            {chatTextStr}
            ''')
            st.write(answer)