app.py

import gradio as gr
import numpy as np
import fitz  # PyMuPDF
import tika
import torch
from fastapi import FastAPI
from transformers import pipeline
from PIL import Image
from io import BytesIO
from starlette.responses import RedirectResponse
from tika import parser
from openpyxl import load_workbook

# Initialize Tika for DOCX & PPTX parsing (Ensure Java is installed)
tika.initVM()

# Initialize FastAPI
app = FastAPI()

# Load models
device = "cuda" if torch.cuda.is_available() else "cpu"
qa_pipeline = pipeline("text-generation", model="TinyLlama/TinyLlama-1.1B-Chat-v1.0", device=device)
image_captioning_pipeline = pipeline("image-to-text", model="nlpconnect/vit-gpt2-image-captioning")

ALLOWED_EXTENSIONS = {"pdf", "docx", "pptx", "xlsx"}

# ✅ Function to Validate File Type
def validate_file_type(file):
    if hasattr(file, "name"):
        ext = file.name.split(".")[-1].lower()
        if ext not in ALLOWED_EXTENSIONS:
            return f"❌ Unsupported file format: {ext}"
        return None
    return "❌ Invalid file format!"

# ✅ Extract Text from PDF
def extract_text_from_pdf(file):
    with fitz.open(file.name) as doc:
        return "\n".join([page.get_text() for page in doc])

# ✅ Extract Text from DOCX & PPTX using Tika
def extract_text_with_tika(file):
    return parser.from_file(file.name)["content"]

# ✅ Extract Text from Excel
def extract_text_from_excel(file):
    wb = load_workbook(file.name, data_only=True)
    text = []
    for sheet in wb.worksheets:
        for row in sheet.iter_rows(values_only=True):
            text.append(" ".join(str(cell) for cell in row if cell))
    return "\n".join(text)

# ✅ Truncate Long Text for Model
def truncate_text(text, max_length=2048):
    return text[:max_length] if len(text) > max_length else text

# ✅ Answer Questions from Image or Document
def answer_question(file, question: str):
    if isinstance(file, np.ndarray):  # Image Processing
        image = Image.fromarray(file)
        caption = image_captioning_pipeline(image)[0]['generated_text']
        response = qa_pipeline(f"Question: {question}\nContext: {caption}")
        return response[0]["generated_text"]

    validation_error = validate_file_type(file)
    if validation_error:
        return validation_error
    
    file_ext = file.name.split(".")[-1].lower()

    # Extract Text from Supported Documents
    if file_ext == "pdf":
        text = extract_text_from_pdf(file)
    elif file_ext in ["docx", "pptx"]:
        text = extract_text_with_tika(file)
    elif file_ext == "xlsx":
        text = extract_text_from_excel(file)
    else:
        return "❌ Unsupported file format!"

    if not text:
        return "⚠️ No text extracted from the document."

    truncated_text = truncate_text(text)
    response = qa_pipeline(f"Question: {question}\nContext: {truncated_text}")

    return response[0]["generated_text"]

# ✅ Gradio Interface (Separate File & Image Inputs)
with gr.Blocks() as demo:
    gr.Markdown("## 📄 AI-Powered Document & Image QA")

    with gr.Row():
        file_input = gr.File(label="Upload Document")
        image_input = gr.Image(label="Upload Image")
    
    question_input = gr.Textbox(label="Ask a Question", placeholder="What is this document about?")
    answer_output = gr.Textbox(label="Answer")
    submit_btn = gr.Button("Get Answer")
    
    submit_btn.click(answer_question, inputs=[file_input, question_input], outputs=answer_output)

# ✅ Mount Gradio with FastAPI
app = gr.mount_gradio_app(app, demo, path="/")

@app.get("/")
def home():
    return RedirectResponse(url="/")