app.py

import gradio as gr
import os
from groq import Groq
from langchain.text_splitter import CharacterTextSplitter
from sentence_transformers import SentenceTransformer
import faiss
from PyPDF2 import PdfReader
from docx import Document
from transformers import pipeline

# Initialize Sentence Transformer for embeddings
model = SentenceTransformer('all-MiniLM-L6-v2')
client = Groq(api_key=os.getenv("groq_api_key"))
# Vector Store (FAISS)
dimension = 384  # Embedding size
index = faiss.IndexFlatL2(dimension)

# Initialize Hugging Face summarization model
summarizer = pipeline("summarization", model="facebook/bart-large-cnn")

# Function to extract text from PDFs
def extract_text_from_pdf(file_path):
    reader = PdfReader(file_path)
    text = ""
    for page in reader.pages:
        text += page.extract_text()
    return text

# Function to extract text from DOCX
def extract_text_from_docx(file_path):
    doc = Document(file_path)
    text = ""
    for paragraph in doc.paragraphs:
        text += paragraph.text + "\n"
    return text

# Function to process files
def process_files(files):
    texts = []
    for file in files:
        if file.name.endswith('.pdf'):
            texts.append(extract_text_from_pdf(file.name))
        elif file.name.endswith('.docx'):
            texts.append(extract_text_from_docx(file.name))
    return texts

# Function to tokenize and chunk text
def chunk_text(text, chunk_size=500, overlap=50):
    text_splitter = CharacterTextSplitter(chunk_size=chunk_size, chunk_overlap=overlap)
    return text_splitter.split_text(text)

# Function to create embeddings and populate FAISS index
def create_embeddings_and_store(chunks):
    global index
    index = faiss.IndexFlatL2(dimension)
    for chunk in chunks:
        embedding = model.encode([chunk])
        embedding = embedding.astype('float32')
        index.add(embedding)

# Function for summarizing the text before sending
def summarize_text(text):
    summary = summarizer(text, max_length=300, min_length=100, do_sample=False)
    return summary[0]['summary_text']

# Function to dynamically truncate context to fit the Groq API's token limit
def truncate_context(context, max_tokens=4000):
    if len(context) > max_tokens:
        context = context[:max_tokens]
    return context

# Function to query Groq with context and question
def query_groq(question, context):
    try:
        if not question.strip():
            return "Error: Question is empty or invalid."
        if not context.strip():
            return "Error: No context available from the uploaded documents."

        max_context_tokens = 4000
        context = truncate_context(context, max_tokens=max_context_tokens)

        chat_completion = client.chat.completions.create(
            messages=[{"role": "system", "content": "You are a helpful assistant. Use the context provided to answer the question."},
                      {"role": "assistant", "content": context},
                      {"role": "user", "content": question}],
            model="llama3-8b-8192", stream=False)
        if chat_completion and chat_completion.choices:
            return chat_completion.choices[0].message.content
        else:
            return "Error: Received an unexpected response from Groq API."
    except Exception as e:
        return f"Error: {str(e)}"

# Function to handle RAG pipeline
def rag_pipeline(files, question, summarize_before_sending=False):
    try:
        if not files:
            return "Error: No files uploaded. Please upload at least one document."

        texts = process_files(files)
        if not texts:
            return "Error: Could not extract text from the uploaded files."

        combined_text = " ".join(texts)

        if summarize_before_sending:
            combined_text = summarize_text(combined_text)

        max_text_size = 4000
        combined_text = truncate_context(combined_text, max_tokens=max_text_size)

        chunks = chunk_text(combined_text)
        create_embeddings_and_store(chunks)

        answer = query_groq(question, combined_text)
        return answer
    except Exception as e:
        return f"Error: {str(e)}"

# Enhanced UI with modern and clean style
with gr.Blocks() as app:
    with gr.Row():
        # Left Column for instructions
        with gr.Column(scale=1, min_width=250):
            gr.Markdown("""
            <div style="background: linear-gradient(145deg, #6e7dff, #1c2b58); padding: 30px; border-radius: 12px; box-shadow: 0 5px 15px rgba(0, 0, 0, 0.1); font-family: 'Roboto', sans-serif;">
                <h2 style="color: #fff; font-size: 32px; font-weight: bold;">DocAI: Document Assistant</h2>
                <p style="color: #ddd; font-size: 18px;">Welcome to DocAI! Upload your documents and get intelligent answers based on their content.</p>
                <p style="color: #ddd; font-size: 16px; line-height: 1.6;"><strong>Steps to use:</strong></p>
                <ul style="color: #ddd; font-size: 16px; line-height: 1.6;">
                    <li>Upload your PDF or DOCX files.</li>
                    <li>Ask questions related to the document.</li>
                    <li>Enable "Summarize Before Sending" for a brief summary of the document.</li>
                    <li>Click "Submit" to get your answers.</li>
                </ul>
                <p style="color: #ddd; font-size: 16px; line-height: 1.6;">Upload multiple files and get answers based on their contents.</p>
            </div>
            """)

        # Right Column for the main application content
        with gr.Column(scale=2, min_width=600):
            gr.Markdown("""
            <div style="background: linear-gradient(135deg, #6e7dff, #1c2b58); padding: 20px; border-radius: 15px; box-shadow: 0 5px 15px rgba(0, 0, 0, 0.2); font-family: 'Roboto', sans-serif;">
                <h2 style="color: #fff; font-size: 36px; font-weight: bold; text-align: center; letter-spacing: 2px; text-transform: uppercase;">
                    Ask Your Document
                </h2>
                <p style="color: #ddd; font-size: 18px; text-align: center; line-height: 1.6;">
                    Get intelligent answers based on the content of your uploaded documents. Just ask a question!
                </p>
            </div>
            """)

            # File input
            file_input = gr.File(
                label="Upload Documents (PDF/DOCX)", 
                file_types=[".pdf", ".docx"], 
                file_count="multiple", 
                interactive=True
            )

            # Question input
            question_input = gr.Textbox(
                label="Ask a question", 
                placeholder="Type your question here...", 
                interactive=True, 
                lines=2, 
                max_lines=4
            )

            # Summarize before sending checkbox
            summarize_before_input = gr.Checkbox(
                label="Summarize Before Sending", 
                value=False
            )

            # Output text box
            output = gr.Textbox(
                label="Answer from LLM", 
                interactive=False, 
                lines=4, 
                max_lines=6
            )

            # Submit button
            submit_button = gr.Button("Submit", icon="send")

            # Apply the logic for the button to trigger the RAG pipeline
            submit_button.click(rag_pipeline, inputs=[file_input, question_input, summarize_before_input], outputs=output)

# Launch the app
app.launch()