app.py

import gradio as gr
import fitz  # PyMuPDF
import torch
from transformers import pipeline
import time, logging, re, difflib
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import io
from PIL import Image

logging.basicConfig(level=logging.ERROR)
device = -1  # CPU-only
print("⚠️ CPU-only. Using faster models!")

# Load faster summarizer
try:
    summarizer = pipeline("summarization", model="sshleifer/distilbart-cnn-12-6", device=device, torch_dtype=torch.float32)
except Exception as e:
    print(f"❌ Summarizer model loading failed: {str(e)}")
    exit(1)

# Load QA model
try:
    qa_pipeline = pipeline("question-answering", model="deepset/roberta-base-squad2", device=device)
except Exception as e:
    print(f"❌ QA model loading failed: {str(e)}")
    exit(1)

def visualize_chunk_status(chunk_data):
    status_colors = {'summarized': 'green', 'skipped': 'orange', 'error': 'red'}
    labels = [f"C{i['chunk']}" for i in chunk_data]
    colors = [status_colors.get(i['status'], 'gray') for i in chunk_data]
    times = [i.get('time', 0.1) for i in chunk_data]

    fig, ax = plt.subplots(figsize=(10, 2.5))
    ax.barh(labels, times, color=colors)
    ax.set_xlabel("Time (s)")
    ax.set_title("📊 Chunk Processing Status")
    plt.tight_layout()

    buf = io.BytesIO()
    plt.savefig(buf, format='png')
    buf.seek(0)
    plt.close(fig)
    return Image.open(buf)

def summarize_file(file_bytes):
    start = time.time()
    chunk_info = []

    try:
        doc = fitz.open(stream=file_bytes, filetype="pdf")
        text = "".join(page.get_text("text") for page in doc)
        text = re.sub(r"\$\s*([^$]+)\s*\$", r"\1", text)
        text = re.sub(r"\\cap", "intersection", text)
        text = re.sub(r"\s+", " ", text).strip()
        text = "".join(c for c in text if ord(c) < 128)
    except Exception as e:
        return f"❌ Text extraction failed: {str(e)}", None

    if not text.strip():
        return "❌ No text found", None

    text = text[:300000]  # allow full but reasonable size
    chunks = [text[i:i+1000] for i in range(0, len(text), 1000)]
    summaries = []

    for i, chunk in enumerate(chunks):
        chunk_start = time.time()
        chunk_result = {'chunk': i + 1, 'status': '', 'time': 0}

        if sum(1 for c in chunk if not c.isalnum()) / len(chunk) > 0.5:
            summaries.append(f"### Chunk {i+1}: Skipped (equation-heavy)")
            chunk_result['status'] = 'skipped'
        else:
            try:
                summary = summarizer(chunk, max_length=150, min_length=50, do_sample=False)[0]['summary_text']
                summaries.append(f"### Chunk {i+1}\n{summary}")
                chunk_result['status'] = 'summarized'
            except Exception as e:
                summaries.append(f"### Chunk {i+1}: ❌ Error: {str(e)}")
                chunk_result['status'] = 'error'

        chunk_result['time'] = time.time() - chunk_start
        chunk_info.append(chunk_result)

    formatted_chunks = "\n\n---\n\n".join(summaries)
    final_summary = f"""**Characters Processed**: {len(text)}
**Total Time**: {time.time()-start:.2f} seconds
## 🔹 Summary by Chunks
{formatted_chunks}
"""
    image = visualize_chunk_status(chunk_info)
    return final_summary, image

def find_relevant_passages(text, question, num_passages=5):
    passages = re.split(r'(?<=[.?!])\s+', text)
    scored = []
    question_tokens = set(question.lower().split())
    for passage in passages:
        passage_tokens = set(passage.lower().split())
        match_score = len(question_tokens.intersection(passage_tokens))
        if match_score == 0:
            match_score = difflib.SequenceMatcher(None, question, passage).ratio()
        scored.append((match_score, passage))
    scored.sort(reverse=True)
    best_passages = " ".join([p for _, p in scored[:num_passages]])
    return best_passages

def answer_question(file_bytes, question):
    try:
        doc = fitz.open(stream=file_bytes, filetype="pdf")
        text = "".join(page.get_text("text") for page in doc)
        text = re.sub(r"\s+", " ", text).strip()
        text = "".join(c for c in text if ord(c) < 128)
        context = text[:300000]
    except Exception as e:
        return f"❌ Text extraction failed: {str(e)}"

    if not question.strip():
        return "⚠️ Please enter a valid question."

    try:
        relevant_context = find_relevant_passages(context, question)
        result = qa_pipeline(question=question, context=relevant_context)
        return f"**Answer**: {result['answer']}\n\n**Score**: {result['score']:.2f}"
    except Exception as e:
        return f"❌ QA failed: {str(e)}"

# Summarizer UI
summarizer_ui = gr.Interface(
    fn=summarize_file,
    inputs=gr.File(label="📄 Upload PDF", type="binary"),
    outputs=[
        gr.Textbox(label="📝 Summarized Output", lines=30, show_copy_button=True),
        gr.Image(label="📊 Visual Process Flow", type="pil")
    ],
    title="📝 AI-Powered PDF Summarizer",
    description="Summarizes long PDFs and visualizes chunk-level processing."
)

# Q&A UI
qa_ui = gr.Interface(
    fn=answer_question,
    inputs=[
        gr.File(label="📄 Upload PDF", type="binary"),
        gr.Textbox(label="❓ Ask a Question")
    ],
    outputs=gr.Textbox(label="🔍 Answer"),
    title="📚 PDF Q&A Assistant",
    description="Ask natural language questions from the uploaded PDF."
)

# Tabs
if __name__ == "__main__":
    try:
        gr.TabbedInterface(
            [summarizer_ui, qa_ui],
            ["📝 Summarizer", "❓ Q&A Assistant"]
        ).launch(server_port=7860)
    except Exception as e:
        print(f"❌ Gradio launch failed: {str(e)}")