app.py

import pdfplumber
from sentence_transformers import SentenceTransformer
import streamlit as st
from gtts import gTTS
import os
from sklearn.metrics.pairwise import cosine_similarity

# Function to extract text from a limited number of pages in a PDF
def extract_text_from_pdf(pdf_path, start_page=0, end_page=10):
    text = ""
    with pdfplumber.open(pdf_path) as pdf:
        for page in pdf.pages[start_page:end_page]:
            text += page.extract_text() + "\n"
    return text

# Load your PDF file (you might want to upload it separately in Spaces)
pdf_path = "/content/Accounting.pdf"  # Update this with the actual file path in Spaces

# Initialize the model
model = SentenceTransformer('all-MiniLM-L6-v2')

# Example: Process the first 100 pages in batches
all_sentences = []
for i in range(0, 300, 10):  # Adjust the step as needed
    pdf_text = extract_text_from_pdf(pdf_path, start_page=i, end_page=i+10)
    all_sentences.extend(pdf_text.split('. '))

# Create embeddings from extracted sentences
pdf_embeddings = model.encode(all_sentences, convert_to_tensor=True)

# Function to respond to user query
def respond_to_query(query):
    query_embedding = model.encode(query, convert_to_tensor=True)
    similarities = cosine_similarity(query_embedding.reshape(1, -1), pdf_embeddings)
    best_match_index = similarities.argmax()
    response = all_sentences[best_match_index]
    return response

# Streamlit app
st.title("Study Assistant")

query = st.text_input("Type your question:")
submit_button = st.button("Ask")

if submit_button:
    if query:
        response = respond_to_query(query)
        
        # Text-to-Speech
        tts = gTTS(response)
        tts.save("response.mp3")
        
        # Playing audio (this might not work in Spaces, consider alternatives)
        os.system("mpg321 response.mp3")
        
        st.write(response)
    else:
        st.write("Please enter a question.")