app.py

import streamlit as st
import fitz  # PyMuPDF for PDF extraction
import faiss
import numpy as np
from transformers import pipeline
from sentence_transformers import SentenceTransformer
import requests
from io import BytesIO
import docx
import pandas as pd

# Initialize the summarization and question-answering models from Hugging Face
summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
qa_pipeline = pipeline("question-answering", model="deepset/roberta-base-squad2")

# Sentence Transformer for embedding-based retrieval
embedder = SentenceTransformer('all-MiniLM-L6-v2')

# FAISS Indexing Function with sentence-level retrieval
def create_faiss_index(text):
    sentences = text.split(". ")  # Split into sentences using "." as delimiter
    embeddings = embedder.encode(sentences)
    index = faiss.IndexFlatL2(384)
    index.add(np.array(embeddings).astype(np.float32))
    return index, sentences

# Function to retrieve the most relevant sentences
def retrieve_relevant_sentences(query, index, sentences):
    query_embedding = embedder.encode([query])
    D, I = index.search(np.array(query_embedding).astype(np.float32), 5)  # Retrieve top 5 most similar sentences
    relevant_sentences = [sentences[i] for i in I[0]]
    return relevant_sentences

# Function to filter retrieved sentences based on keywords
def filter_sentences(query, sentences):
    filtered_sentences = []
    for sentence in sentences:
        if any(word.lower() in sentence.lower() for word in query.split()):
            filtered_sentences.append(sentence)
    return filtered_sentences

# Streamlit UI
st.title("Concise Summarizer and Q&A")

# Upload File
uploaded_file = st.file_uploader("Upload a PDF, Word, or Excel file", type=["pdf", "docx", "xlsx"])

if uploaded_file:
    file_type = uploaded_file.type

    # Extract text based on file type
    if file_type == "application/pdf":
        doc = fitz.open(stream=uploaded_file.read(), filetype="pdf")
        text = ""
        for page in doc:
            text += page.get_text()
    elif file_type == "application/vnd.openxmlformats-officedocument.wordprocessingml.document":
        doc = docx.Document(BytesIO(uploaded_file.read()))
        text = ""
        for para in doc.paragraphs:
            text += para.text + "\n"
    elif file_type == "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet":
        df = pd.read_excel(uploaded_file, engine="openpyxl")
        text = ""
        for col in df.columns:
            text += "\n".join(df[col].dropna().astype(str).tolist()) + "\n"
    else:
        st.error("Unsupported file type!")
        text = ""

    if text:
        # Display the first 500 characters of extracted text
        st.write("Text extracted from file:")
        st.write(text[:500])  # Show first 500 characters

        # Create FAISS index
        index, sentences = create_faiss_index(text)

        # Input for user query
        query = st.text_input("Enter your query:")

        if query:
            st.write("Retrieving relevant information...")
            relevant_sentences = retrieve_relevant_sentences(query, index, sentences)
            filtered_sentences = filter_sentences(query, relevant_sentences)

            # Combine filtered sentences into a single string
            relevant_text = " ".join(filtered_sentences)

            st.write(f"Relevant Text: {relevant_text}")

            # Answer the question based on the relevant chunk
            st.write("Answering the question...")
            try:
                answer = qa_pipeline(question=query, context=relevant_text)
                concise_answer = answer['answer']
                st.write(f"Answer: {concise_answer}")
            except Exception as e:
                st.write(f"Error answering question: {str(e)}")

            # Summarize the relevant chunk (concise summary after query answer)
            if relevant_text.strip():
                if len(relevant_text.split()) > 20:  # Only summarize if text is long enough
                    try:
                        st.write("Summarizing...")
                        summary = summarizer(relevant_text, max_length=50, min_length=30, do_sample=False)[0]['summary_text']
                        st.write(f"Summary: {summary}")
                    except Exception as e:
                        st.write(f"Error summarizing text: {str(e)}")
                else:
                    st.write("Text is too short to summarize.")
            else:
                st.write("No relevant text found to summarize.")