app.py

import os
import torch
import gradio as gr
import pdfplumber
import docx
from transformers import pipeline

# Load Hugging Face token from environment
hf_token = os.environ.get("HF_TOKEN")

# Model ID (Gemma Instruct)
model_id = "google/gemma-7b-it"

# Detect device
device = 0 if torch.cuda.is_available() else -1
torch_dtype = torch.float16 if device == 0 else torch.float32

# Load the instruction-following pipeline
pipe = pipeline("text-generation",
                model=model_id,
                tokenizer=model_id,
                use_auth_token=hf_token,
                device=device,
                torch_dtype=torch_dtype,
                max_new_tokens=1024)

# Extract text from PDF
def extract_text_from_pdf(file):
    text = ""
    with pdfplumber.open(file.name) as pdf:
        for page in pdf.pages:
            page_text = page.extract_text()
            if page_text:
                text += page_text + "\n"
    return text

# Extract text from DOCX
def extract_text_from_docx(file):
    doc = docx.Document(file)
    return "\n".join([para.text for para in doc.paragraphs if para.text.strip()])

# Summarize document contents
def summarize_document(file):
    if file.name.endswith(".pdf"):
        full_text = extract_text_from_pdf(file)
    elif file.name.endswith(".docx"):
        full_text = extract_text_from_docx(file)
    else:
        return "❌ Please upload a PDF or DOCX file."

    chunks = [full_text[i:i+1500] for i in range(0, len(full_text), 1500)]
    summary = ""

    for i, chunk in enumerate(chunks[:3]):
        prompt = f"""Read the following technical/tender document chunk and extract these key points:
        
1. Number of workers or people required  
2. Timeline or duration for project completion  
3. Technologies, tools, or machines mentioned  
4. Any deadlines, conditions, or legal terms  

Document chunk:
{chunk}

Please return only the extracted information in clean bullet points."""
        result = pipe(prompt)[0]["generated_text"]
        summary += f"\n\n--- Chunk {i+1} ---\n" + result.split("Document chunk:")[-1].strip()

    return summary.strip()

# Gradio Interface
iface = gr.Interface(
    fn=summarize_document,
    inputs=gr.File(label="📄 Upload Tender Document (PDF or DOCX)"),
    outputs=gr.Textbox(label="🧾 Extracted Summary", lines=30),
    title="📘 Smart Tender Analyzer (Gemma-7B)",
    description="Upload a tender or technical document (PDF/DOCX). This app extracts important project info using Google's Gemma-7B."
)

# Launch app (no share=True for Hugging Face Spaces)
iface.launch()