src/streamlit_app.py

import streamlit as st
import pandas as pd
import pytesseract
from PIL import Image
from rapidfuzz import fuzz, utils
import io
import easyocr
import numpy as np

# Initialize the reader (do this outside the function so it only loads once)
reader = easyocr.Reader(['en']) 

# --- CONFIGURATION ---
st.set_page_config(page_title="AI Student Grader", layout="wide")
st.title("📝 AI Student Answer Grader")
st.markdown("Upload answer sheets and an answer key to automatically calculate marks.")

# --- SIDEBAR SETTINGS ---
st.sidebar.header("Grading Settings")
accuracy_threshold = st.sidebar.slider("Minimum Accuracy Threshold (%)", 0, 100, 70)
marks_per_question = st.sidebar.number_input("Marks per correct answer", value=1.0)

# --- HELPER FUNCTIONS ---
def perform_ocr(image):
    """Extracts text using EasyOCR."""
    img = Image.open(image)
    # Convert PIL image to numpy array for EasyOCR
    img_array = np.array(img)
    results = reader.readtext(img_array, detail=0) # detail=0 returns only text
    return "\n".join(results)

def compare_answers(student_text, answer_key, threshold):
    """
    Compares student text with answer key using Fuzzy Matching.
    DeepSeek-R1 style logic: We look for the presence of key concepts.
    """
    # Simple line-by-line comparison (assuming 1 question per line or similar structure)
    # For complex papers, you'd split by question numbers
    score = 0
    key_lines = [line.strip() for line in answer_key.split('\n') if line.strip()]
    student_lines = [line.strip() for line in student_text.split('\n') if line.strip()]
    
    details = []
    
    for i, correct_ans in enumerate(key_lines):
        match_found = False
        highest_match = 0
        
        # Compare against each line in student text to find the best match for this answer
        for s_line in student_lines:
            similarity = fuzz.token_set_ratio(correct_ans, s_line)
            if similarity > highest_match:
                highest_match = similarity
        
        if highest_match >= threshold:
            score += marks_per_question
            match_found = True
            
        details.append({
            "Question": i + 1,
            "Match %": round(highest_match, 2),
            "Status": "Correct" if match_found else "Incorrect"
        })
            
    return score, details

# --- UI LAYOUT ---
col1, col2 = st.columns(2)

with col1:
    st.subheader("1. Reference Answer Key")
    key_input_type = st.radio("Key Format", ["Text Input", "Upload Image"])
    
    if key_input_type == "Text Input":
        answer_key_text = st.text_area("Paste the correct answers (one per line):")
    else:
        key_img = st.file_uploader("Upload Answer Key Image", type=['png', 'jpg', 'jpeg'])
        if key_img:
            answer_key_text = perform_ocr(key_img)
            st.text_area("Extracted Key (Edit if needed):", value=answer_key_text)

with col2:
    st.subheader("2. Student Answer Sheets")
    student_images = st.file_uploader("Upload Student Images (Max 5)", type=['png', 'jpg', 'jpeg'], accept_multiple_files=True)

# --- PROCESSING ---
if st.button("Calculate Marks"):
    if not answer_key_text or not student_images:
        st.error("Please provide both the answer key and student images.")
    else:
        results = []
        
        progress_bar = st.progress(0)
        for idx, img_file in enumerate(student_images):
            # 1. OCR
            extracted_text = perform_ocr(img_file)
            
            # 2. Compare
            score, details = compare_answers(extracted_text, answer_key_text, accuracy_threshold)
            
            # 3. Store Results
            results.append({
                "Student Name": img_file.name.split('.')[0], # Uses filename as name
                "Raw Score": score,
                "Final Marks": f"{score}/{len(answer_key_text.splitlines()) * marks_per_question}",
                "Match Percentage": f"{accuracy_threshold}%"
            })
            progress_bar.progress((idx + 1) / len(student_images))

        # --- DISPLAY RESULTS ---
        df = pd.DataFrame(results)
        st.subheader("📊 Results Overview")
        st.table(df)

        # --- EXCEL EXPORT ---
        output = io.BytesIO()
        with pd.ExcelWriter(output, engine='openpyxl') as writer:
            df.to_excel(writer, index=False, sheet_name='Grades')
        
        st.download_button(
            label="📥 Download Excel Sheet",
            data=output.getvalue(),
            file_name="student_grades.xlsx",
            mime="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet"
        )