app.py

import streamlit as st
import openai
import fitz  # PyMuPDF
import numpy as np
from sklearn.metrics.pairwise import cosine_similarity
from io import BytesIO

# Function to extract text from the uploaded PDF file
def extract_pdf_text(pdf_file):
    doc = fitz.open(stream=pdf_file.read(), filetype="pdf")
    text = ""
    for page in doc:
        text += page.get_text("text")
    return text

# Function to get embeddings for the text
def get_embeddings(texts):
    response = openai.Embedding.create(
        model="text-embedding-ada-002",
        input=texts
    )
    embeddings = [embedding['embedding'] for embedding in response['data']]
    return embeddings

# Function to get the most relevant context from the PDF for the query
def get_relevant_context(pdf_text, query, num_contexts=3):
    # Split the PDF text into chunks for better matching
    pdf_text_chunks = [pdf_text[i:i+1500] for i in range(0, len(pdf_text), 1500)]
    # Get embeddings for both the document and the query
    pdf_embeddings = get_embeddings(pdf_text_chunks)
    query_embedding = get_embeddings([query])[0]
    
    # Compute cosine similarity between query and document chunks
    similarities = cosine_similarity([query_embedding], pdf_embeddings)
    top_indices = similarities[0].argsort()[-num_contexts:][::-1]
    
    # Combine the top context pieces
    relevant_context = " ".join([pdf_text_chunks[i] for i in top_indices])
    return relevant_context

# Function to generate a response from GPT-4 chat model
def generate_response(context, question):
    messages = [
        {"role": "system", "content": "You are a helpful assistant expert on GPT-4."},
        {"role": "user", "content": f"Context: {context}\nQuestion: {question}"}
    ]
    response = openai.ChatCompletion.create(
        model="gpt-4o-mini",  # Use the GPT-4 chat model
        messages=messages,
        max_tokens=1200,
        temperature=0.7,
    )
    return response['choices'][0]['message']['content'].strip()

# Function to handle irrelevant questions
def is_irrelevant_question(question):
    irrelevant_keywords = ["life", "love", "meaning", "future", "philosophy"]
    return any(keyword in question.lower() for keyword in irrelevant_keywords)

# Streamlit UI
def main():
    st.title("📄 GPT-4 Research Paper Chatbot")
    st.write("💬 Ask any question related to the GPT-4 paper, and I'll try to answer it!")
    
    # User input: OpenAI API key
    openai_api_key = st.text_input("🔑 Enter your OpenAI API Key:", type="password")
    
    if openai_api_key:
        openai.api_key = openai_api_key
        st.success("API Key successfully set!")

        # Upload the PDF file
        pdf_file = st.file_uploader("📂 Upload GPT-4 Research Paper PDF", type="pdf")
        
        if pdf_file is not None:
            # Extract text from the uploaded PDF
            pdf_text = extract_pdf_text(pdf_file)
            st.write("✅ PDF content loaded successfully! Start asking questions.")

            # User input: the question they want to ask
            question = st.text_input("Ask your question:")

            if question:
                # Check if the question is irrelevant
                if is_irrelevant_question(question):
                    st.write("Sorry, I don't know the answer to this question. I am an expert on GPT-4 knowledge.")
                else:
                    # Get the most relevant context from the document
                    relevant_context = get_relevant_context(pdf_text, question)
                    
                    # Generate the response from GPT-4 chat model
                    answer = generate_response(relevant_context, question)
                    
                    # Display the answer
                    st.write(f"🤖 Answer: {answer}")
    else:
        st.warning("⚠️ Please enter your OpenAI API Key to use the chatbot.")

if __name__ == "__main__":
    main()