app.py

import gradio as gr
from langchain.document_loaders import PyPDFLoader
from langchain.text_splitter import CharacterTextSplitter
from langchain.embeddings import SentenceTransformerEmbeddings
from langchain.vectorstores import FAISS
from langchain.memory import ConversationBufferMemory
from groq import Groq
import requests
from bs4 import BeautifulSoup

client = Groq(api_key="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")
embedding_model = SentenceTransformerEmbeddings(model_name="sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2")

memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)

def process_pdf_with_langchain(pdf_path):
    """Process the PDF file using LangChain for RAG."""
    loader = PyPDFLoader(pdf_path)
    documents = loader.load()
    text_splitter = CharacterTextSplitter(chunk_size=500, chunk_overlap=50)
    split_documents = text_splitter.split_documents(documents)

    vectorstore = FAISS.from_documents(split_documents, embedding_model)
    retriever = vectorstore.as_retriever(search_kwargs={"k": 3})
    return retriever

def scrape_google_search(query, num_results=3):
    """Search Google and return the top results."""
    headers = {"User-Agent": "Mozilla/5.0"}
    search_url = f"https://www.google.com/search?q={query}"
    response = requests.get(search_url, headers=headers)
    soup = BeautifulSoup(response.text, "html.parser")

    results = []
    for g in soup.find_all('div', class_='tF2Cxc')[:num_results]:
        title = g.find('h3').text
        link = g.find('a')['href']
        results.append(f"{title}: {link}")
    return "\n".join(results)

def generate_response(query, retriever=None, use_web_search=False):
    """Generate a response using LangChain with optional retriever and web search."""
    knowledge = ""

    if retriever:
        relevant_docs = retriever.get_relevant_documents(query)
        knowledge += "\n".join([doc.page_content for doc in relevant_docs])

    if use_web_search:
        web_results = scrape_google_search(query)
        knowledge += f"\n\nWeb Search Results:\n{web_results}"

    chat_history = memory.load_memory_variables({}).get("chat_history", "")
    context = (
        f"This is a conversation with ParvizGPT, an AI model designed by Amir Mahdi Parviz, "
        f"to help with tasks like answering questions in Persian, providing recommendations, and decision-making."
    )
    if knowledge:
        context += f"\n\nRelevant Knowledge:\n{knowledge}"
    if chat_history:
        context += f"\n\nChat History:\n{chat_history}"

    context += f"\n\nYou: {query}\nParvizGPT:"

    chat_completion = client.chat.completions.create(
        messages=[{"role": "user", "content": context}],
        model="llama-3.3-70b-versatile",
    )
    response = chat_completion.choices[0].message.content.strip()

    memory.save_context({"input": query}, {"output": response})
    return response

def gradio_interface(user_message, pdf_file=None, enable_web_search=False):
    global retriever
    if pdf_file is not None:
        try:
            retriever = process_pdf_with_langchain(pdf_file.name)
        except Exception as e:
            return f"Error processing PDF: {e}"

    response = generate_response(user_message, retriever=retriever, use_web_search=enable_web_search)
    return response

def clear_memory():
    memory.clear()
    return "Memory cleared!"

retriever = None  

with gr.Blocks() as interface:
    gr.Markdown("## ParvizGPT with Memory and Web Search Toggle")
    with gr.Row():
        user_message = gr.Textbox(label="Your Question", placeholder="Type your question here...", lines=1)
        submit_btn = gr.Button("Submit")
    with gr.Row():
        pdf_file = gr.File(label="Upload PDF for Context (Optional)", type="filepath")
        enable_web_search = gr.Checkbox(label="Enable Web Search", value=False)
    with gr.Row():
        clear_memory_btn = gr.Button("Clear Memory")
    response_output = gr.Textbox(label="Response", placeholder="ParvizGPT's response will appear here.")

    submit_btn.click(gradio_interface, inputs=[user_message, pdf_file, enable_web_search], outputs=response_output)
    clear_memory_btn.click(clear_memory, inputs=[], outputs=response_output)

    gr.HTML(
        """
        <script>
        document.addEventListener("keydown", function(event) {
            if (event.key === "Enter" && !event.shiftKey) {
                event.preventDefault();
                document.querySelector('button[title="Submit"]').click();
            }
        });
        </script>
        """
    )

interface.launch()