app.py

import gradio as gr
from PIL import Image
import pytesseract
import cv2
import numpy as np
from transformers import pipeline
import os
print("Files in current directory:", os.listdir())

import csv

# Set transformers cache to writable directory
os.environ["TRANSFORMERS_CACHE"] = "/app/.cache/huggingface"

# Load the LLM pipeline (using a smaller model for easier local use)
llm = pipeline(
    "text-generation",
    model="tiiuae/falcon-rw-1b",
    tokenizer="tiiuae/falcon-rw-1b",
    max_new_tokens=300
)

# Function to parse CSV and format plans as text using your CSV headers
def format_singtel_plans_from_csv(csv_path):
    formatted_plans = []
    with open(csv_path, 'r', encoding='utf-8') as file:
        reader = csv.DictReader(file)
        for row in reader:
            plan_str = f"- {row['Plan Name']}: {row['Badge']}, {row['Price']}, {row['Features']}"
            formatted_plans.append(plan_str)
    return "\n".join(formatted_plans)

# Prepare the plan text once at startup
csv_path = os.path.join(os.path.dirname(__file__), "final_singtel_postpaid_mobile_plans.csv")
plan_text = format_singtel_plans_from_csv(csv_path)

def ocr_extract_and_recommend(image):
    if image is None:
        return "Please upload an image."
    
    # OCR preprocessing
    image_np = np.array(image)
    image_cv = cv2.cvtColor(image_np, cv2.COLOR_RGB2BGR)
    gray = cv2.cvtColor(image_cv, cv2.COLOR_BGR2GRAY)
    _, thresh = cv2.threshold(gray, 150, 255, cv2.THRESH_BINARY)

    # Extract text using pytesseract
    text = pytesseract.image_to_string(thresh).strip()
    if not text:
        return "Couldn't extract any text. Please upload a clearer image."

    # Construct prompt for LLM
    prompt = f"""
You are a telco advisor helping users compare mobile plans.

The user's current mobile bill contains:
{text}

Available Singtel mobile plans:
{plan_text}

Recommend the best Singtel plan (if applicable) and explain why it is better in terms of price, data, and features.
"""

    # Get recommendation from LLM
    try:
        response = llm(prompt)[0]["generated_text"]
    except Exception as e:
        response = f"LLM generation failed: {str(e)}"

    return f"📄 **Extracted Bill Info:**\n{text}\n\n📢 **Recommendation:**\n{response}"

# Gradio interface
iface = gr.Interface(
    fn=ocr_extract_and_recommend,
    inputs=gr.Image(type="pil", label="Upload your bill"),
    outputs="text",
    title="Singtel Plan Recommender with OCR",
    description="Upload your telco bill image to extract plan details and get personalized Singtel plan recommendations."
)

if __name__ == "__main__":
    iface.launch(server_name="0.0.0.0", server_port=7860)