app.py

import streamlit as st
import tempfile
import logging
from typing import List
from langchain.document_loaders import PyPDFLoader
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.vectorstores import FAISS
from langchain.llms import HuggingFacePipeline
from langchain.chains.summarize import load_summarize_chain
from langchain.schema import Document
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.prompts import PromptTemplate
from transformers import pipeline

# Set up logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Constants
EMBEDDING_MODEL = 'sentence-transformers/all-MiniLM-L6-v2'
DEFAULT_MODEL = "facebook/bart-large-cnn"

@st.cache_resource
def load_embeddings():
    """Load and cache the embedding model."""
    try:
        return HuggingFaceEmbeddings(model_name=EMBEDDING_MODEL)
    except Exception as e:
        logger.error(f"Failed to load embeddings: {e}")
        st.error("Failed to load the embedding model. Please try again later.")
        return None

@st.cache_resource
def load_llm(model_name):
    """Load and cache the language model."""
    try:
        pipe = pipeline("text2text-generation", model=model_name, max_length=1024)
        return HuggingFacePipeline(pipeline=pipe)
    except Exception as e:
        logger.error(f"Failed to load LLM: {e}")
        st.error(f"Failed to load the model {model_name}. Please try again.")
        return None

def process_pdf(file) -> List[Document]:
    """Process the uploaded PDF file."""
    try:
        with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as temp_file:
            temp_file.write(file.getvalue())
            temp_file_path = temp_file.name
        
        loader = PyPDFLoader(file_path=temp_file_path)
        pages = loader.load()
        text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
        documents = text_splitter.split_documents(pages)
        return documents
    except Exception as e:
        logger.error(f"Error processing PDF: {e}")
        st.error("Failed to process the PDF. Please make sure it's a valid PDF file.")
        return []

def create_vector_store(documents: List[Document], embeddings):
    """Create the vector store."""
    try:
        return FAISS.from_documents(documents, embeddings)
    except Exception as e:
        logger.error(f"Error creating vector store: {e}")
        st.error("Failed to create the vector store. Please try again.")
        return None

def summarize_report(documents: List[Document], llm) -> str:
    """Summarize the report using the loaded model."""
    try:
        prompt_template = """
        You are an AI specialized in summarizing comprehensive reports with a focus on funding, finances, and global comparisons. Given the detailed report content below, generate a concise and structured summary using bullet points and emojis. The summary should highlight key funding figures, financial data, budget allocations, comparisons between regions, and notable insights about [FOCUS_REGION]'s role in the global context of [TOPIC].
        Report Content:
        {text}
        Your summary should follow this structure:
        Summary:
        💰 [TOPIC] Overview for [FOCUS_REGION]:
       
        🔴 [FOCUS_REGION]'s Position in Global [TOPIC]:
        📍 Total investment/funding: [amount]
        📍 Breakdown of funding sources (e.g., government, private sector)
        📍 [FOCUS_REGION]'s ranking in global investment
        📍 Key statistics and figures
        
        🔴 Financial Impact and Projections:
        📍 Expected ROI or economic benefits
        📍 Financial milestones or targets
        📍 Impact on relevant areas
        
        🔴 Global Comparison:
        📍 [List of relevant countries/regions with their financial figures]
        📍 Comparative analysis of [FOCUS_REGION] vs other major players
        
        🔴 Budget Analysis:
        📍 Major budget items
        📍 Key budget allocations
        📍 Year-over-year budget changes
        📍 Comparison to industry benchmarks
        
        🔴 Funding Strategies:
        📍 Key funding mechanisms (e.g., grants, loans, public-private partnerships)
        📍 Innovative financing approaches
        
        🔴 Progress and Significance:
        📍 Key achievements or milestones
        📍 [1-2 concluding points about [FOCUS_REGION]'s role or significance in [TOPIC]]
        
        Please ensure the summary is concise, informative, and easy to read at a glance. Use precise figures where available and highlight any significant financial trends or insights. The summary should provide a comprehensive overview of both the financial aspects and the broader context of [TOPIC] in [FOCUS_REGION].
        """

        prompt = PromptTemplate.from_template(prompt_template)
        chain = load_summarize_chain(llm, chain_type="stuff", prompt=prompt)
        summary = chain.invoke(documents)
        return summary['output_text']

    except Exception as e:
        logger.error(f"Error summarizing report: {e}")
        st.error("Failed to summarize the report. Please try again.")
        return ""

def main():
    st.title("Report Summarizer")
    
    model_option = st.sidebar.text_input("Enter model name", value=DEFAULT_MODEL)

    uploaded_file = st.sidebar.file_uploader("Upload your Report", type="pdf")

    llm = load_llm(model_option)
    embeddings = load_embeddings()

    if not llm or not embeddings:
        return

    if uploaded_file:
        with st.spinner("Processing PDF..."):
            documents = process_pdf(uploaded_file)

        if documents:
            with st.spinner("Creating vector store..."):
                db = create_vector_store(documents, embeddings)

            if db and st.button("Summarize"):
                with st.spinner(f"Generating structured summary using {model_option}..."):
                    summary = summarize_report(documents, llm)

                    if summary:
                        st.subheader("Structured Summary:")
                        st.markdown(summary)
                    else:
                        st.warning("Failed to generate summary. Please try again.")

if __name__ == "__main__":
    main()