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
@st.cache_resource
def load_pdf_and_extract_text(pdf_path, max_pages=20):
    all_sentences = []
    with pdfplumber.open(pdf_path) as pdf:
        total_pages = len(pdf.pages)
        st.write(f"Total pages to process: {total_pages}")
        for i, page in enumerate(pdf.pages):
            if i >= max_pages:
                break
            st.write(f"Processing page {i + 1}...")
            text = page.extract_text()
            if text:
                all_sentences.extend(text.split('. '))
            st.progress((i + 1) / max_pages)  # Update progress
    return all_sentences

# Load your PDF file
pdf_path = "Accounting.pdf"  # Ensure this is uploaded to your space
all_sentences = load_pdf_and_extract_text(pdf_path)

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

# Create embeddings from extracted sentences
@st.cache_resource
def create_embeddings(sentences):
    return model.encode(sentences, convert_to_tensor=True)

pdf_embeddings = create_embeddings(all_sentences)

# 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")
        
        # (Optional) Playing audio might not work in Spaces, consider alternatives
        st.write(response)
    else:
        st.write("Please enter a question.")