import gradio as gr
from transformers import pipeline
import pytesseract
from sentence_transformers import SentenceTransformer, util
from PIL import Image
from typing import List
import requests

# Initialize sentence transformer model
model1 = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')

# Hugging Face API details
API_URL = "https://api-inference.huggingface.co/models/openai-community/gpt2"
headers = {"Authorization": "Bearer hf_TsCTtXxnvpmhFKABqKmcVLyLEhjQPsITSVx"}

# Function to interact with Hugging Face API for GPT-2
def query(payload):
    response = requests.post(API_URL, headers=headers, json=payload)
    return response.json()

# Function to generate text response from GPT-2 model using Hugging Face API
def generate_response(prompt):
    response = query({"inputs": prompt})
    
    # Check if the response contains the expected format
    if isinstance(response, list) and len(response) > 0 and 'generated_text' in response[0]:
        return response[0]['generated_text']
    else:
        # Log the response if something unexpected is returned
        print("Unexpected response format:", response)
        return "Sorry, I couldn't generate a response."
    

# Function to generate text response from GPT-2 model using Hugging Face API
def generate_response(prompt):
    response = query({"inputs": prompt})
    return response[0]['generated_text']

# Extract text from an image using Tesseract
def extract_text_from_image(filepath: str, languages: List[str]):
    image = Image.open(filepath)
    lang_str = '+'.join(languages)  # Join languages for Tesseract
    return pytesseract.image_to_string(image=image, lang=lang_str)

# Function to get embeddings for text using SentenceTransformer
def get_embedding(text):
    return model1.encode(text, convert_to_tensor=True)

# Calculate similarity between two texts using cosine similarity
def calculate_similarity(text1, text2):
    embedding1 = get_embedding(text1)
    embedding2 = get_embedding(text2)
    similarity = util.pytorch_cos_sim(embedding1, embedding2)
    return similarity.item()

# Assign grades based on similarity score
def get_grade(similarity_score):
    if similarity_score >= 0.9:
        return 5
    elif similarity_score >= 0.8:
        return 4
    elif similarity_score >= 0.7:
        return 3
    elif similarity_score >= 0.6:
        return 2
    else:
        return 1

# Function to evaluate student's answer by comparing it to a model answer
def evaluate_answer(image, languages):
    student_answer = extract_text_from_image(image, languages)
    model_answer = "The process of photosynthesis helps plants produce glucose using sunlight."
    similarity_score = calculate_similarity(student_answer, model_answer)
    grade = get_grade(similarity_score)
    feedback = f"Student's answer: {student_answer}\nTeacher's answer: {model_answer}"
    return grade, similarity_score * 100, feedback

# Main interface function for Gradio
def gradio_interface(image, languages: List[str], prompt):
    grade, similarity_score, feedback = evaluate_answer(image, languages)
    response = generate_response(prompt)
    return grade, similarity_score, feedback, response

# Get available Tesseract languages
language_choices = pytesseract.get_languages()

# Define Gradio interface
interface = gr.Interface(
    fn=gradio_interface,
    inputs=[
        gr.Image(type="filepath", label="Input"), 
        gr.CheckboxGroup(language_choices, type="value", value=['eng'], label='language'),
        gr.Textbox(lines=2, placeholder="Enter your prompt here", label="Prompt")
    ],
    outputs=[
        gr.Text(label="Grade"), 
        gr.Number(label="Similarity Score (%)"), 
        gr.Text(label="Feedback"), 
        gr.Text(label="Generated Response")
    ],
    title="Automated Grading System",
    description="Upload an image of your answer sheet to get a grade from 1 to 5, similarity score, and feedback based on the model answer.",
    live=True
)

if __name__ == "__main__":
    interface.launch()