app.py

import streamlit as st
from langchain.prompts import PromptTemplate
from langchain.chains import LLMChain
from langchain_google_genai import ChatGoogleGenerativeAI
import fitz
import json
import docx
import os


# Session states
if "mcqs" not in st.session_state:
    st.session_state.mcqs = []
if "current_q" not in st.session_state:
    st.session_state.current_q = 0
if "user_answers" not in st.session_state:
    st.session_state.user_answers = {}
if "quiz_finished" not in st.session_state:
    st.session_state.quiz_finished = False
if "language" not in st.session_state:
    st.session_state.language = "English"

# Language selection in sidebar
st.sidebar.title("🌐 Language / 언어")
language = st.sidebar.selectbox(
    "Select Language / 언어 선택",
    ["English", "한국어"],
    index=0 if st.session_state.language == "English" else 1
)
st.session_state.language = language

# UI Text Dictionary
ui_text = {
    "English": {
        "title": "📄 PDF/Word based Self EXAM",
        "sidebar_title": "Upload & Settings",
        "upload_prompt": "Upload a file (PDF or Word)",
        "num_questions": "Number of questions",
        "generate_button": "Generate EXAMs",
        "no_file_error": "Please upload a file.",
        "generating": "Extracting text and generating EXAMs...",
        "success": "✅ EXAMs generated successfully!",
        "error": "Error generating EXAMs:",
        "question_prefix": "Question",
        "choose_answer": "Choose an answer:",
        "next_button": "Next",
        "quiz_completed": "🎉 Quiz completed!",
        "results_header": "📊 Quiz Results",
        "your_answer": "Your answer:",
        "correct_answer": "Correct answer:",
        "score": "✅ You scored {score} out of {total}"
    },
    "한국어": {
        "title": "📄 PDF/Word 기반 객관식 문제 생성기",
        "sidebar_title": "파일 업로드 및 설정",
        "upload_prompt": "파일 업로드 (PDF 또는 Word)",
        "num_questions": "문제 개수",
        "generate_button": "문제 생성",
        "no_file_error": "파일을 업로드해주세요.",
        "generating": "텍스트 추출 및 문제 생성 중...",
        "success": "✅ 문제가 성공적으로 생성되었습니다!",
        "error": "문제 생성 오류:",
        "question_prefix": "문제",
        "choose_answer": "답을 선택하세요:",
        "next_button": "다음",
        "quiz_completed": "🎉 퀴즈가 완료되었습니다!",
        "results_header": "📊 퀴즈 결과",
        "your_answer": "당신의 답:",
        "correct_answer": "정답:",
        "score": "✅ {total}문제 중 {score}문제를 맞추셨습니다"
    }
}

# Get current language texts
texts = ui_text[language]

# Title
st.title(texts["title"])

# Sidebar
st.sidebar.title(texts["sidebar_title"])

# Upload file
uploaded_file = st.sidebar.file_uploader(texts["upload_prompt"], type=["pdf", "docx"])

# Number of questions
number_of_questions = st.sidebar.slider(texts["num_questions"], min_value=1, max_value=20, value=5)

# Gemini setup
GOOGLE_API_KEY = os.getenv("GOOGLE_API_KEY")
llm = ChatGoogleGenerativeAI(
    model="gemini-2.0-flash-exp",
    google_api_key=GOOGLE_API_KEY,
    temperature=0.7
)

# Templates for different languages
template_english = ("You are an expert EXAM generator. Generate {number} unique multiple-choice questions from the given text.\n"
                   "Each question must have exactly 1 correct answer and 3 incorrect options.\n"
                   "Strictly return output in the following JSON format (no explanations, no markdown):\n"
                   "[\n"
                   "  {{\n"
                   '    "question": "What is ...?",\n'
                   '    "options": ["Option A", "Option B", "Option C", "Option D"],\n'
                   '    "answer": "Option D"\n'
                   "  }},\n"
                   "  ...\n"
                   "]\n"
                   "TEXT:\n"
                   "{text}")

template_korean = ("당신은 전문 객관식 문제 출제자입니다. 주어진 텍스트에서 {number}개의 고유한 객관식 문제를 생성하세요.\n"
                  "각 문제는 정확히 1개의 정답과 3개의 오답을 가져야 합니다.\n"
                  "다음 JSON 형식으로만 출력하세요 (설명이나 마크다운 없이):\n"
                  "[\n"
                  "  {{\n"
                  '    "question": "...는 무엇입니까?",\n'
                  '    "options": ["선택지 A", "선택지 B", "선택지 C", "선택지 D"],\n'
                  '    "answer": "선택지 D"\n'
                  "  }},\n"
                  "  ...\n"
                  "]\n"
                  "텍스트:\n"
                  "{text}")

# Select template based on language
template = template_english if language == "English" else template_korean

prompt = PromptTemplate(
    input_variables=["text", "number"],
    template=template
)

mcq_chain = LLMChain(llm=llm, prompt=prompt)

# Extract text from PDF or Word
def extract_text(file):
    if file.name.endswith(".pdf"):
        # Read the entire file content into memory
        file_bytes = file.read()
        # Open the PDF from the byte stream
        doc = fitz.open(stream=file_bytes, filetype="pdf")
        # Extract text from all pages
        text = ""
        for page in doc:
            text += page.get_text()
        return text
    elif file.name.endswith(".docx"):
        doc = docx.Document(file)
        return "\n".join([para.text for para in doc.paragraphs])
    return ""

# Generate MCQs
if st.sidebar.button(texts["generate_button"]):
    if uploaded_file is None:
        st.error(texts["no_file_error"])
    else:
        with st.spinner(texts["generating"]):
            text = extract_text(uploaded_file)
            try:
                response = mcq_chain.run(text=text, number=str(number_of_questions))
                # Clean the response to extract JSON
                response = response.strip()
                if response.startswith("```json"):
                    response = response[7:]
                if response.endswith("```"):
                    response = response[:-3]
                mcqs_json = json.loads(response)
                st.session_state.mcqs = mcqs_json
                st.session_state.current_q = 0
                st.session_state.user_answers = {}
                st.session_state.quiz_finished = False
                st.success(texts["success"])
            except Exception as e:
                st.error(f"{texts['error']} {e}")

# Display question
if st.session_state.mcqs and not st.session_state.quiz_finished:
    idx = st.session_state.current_q
    q_data = st.session_state.mcqs[idx]

    st.subheader(f"{texts['question_prefix']} {idx + 1}: {q_data['question']}")
    
    with st.form(key=f"form_{idx}"):
        selected_option = st.radio(texts["choose_answer"], q_data["options"], key=f"radio_{idx}")
        submitted = st.form_submit_button(texts["next_button"])

        if submitted:
            st.session_state.user_answers[idx] = selected_option
            if idx < len(st.session_state.mcqs) - 1:
                st.session_state.current_q += 1
                st.rerun()
            else:
                st.session_state.quiz_finished = True
                st.success(texts["quiz_completed"])
                st.rerun()

# Show result
if st.session_state.quiz_finished:
    st.header(texts["results_header"])
    score = 0
    total = len(st.session_state.mcqs)

    for i, q in enumerate(st.session_state.mcqs):
        user_ans = st.session_state.user_answers.get(i)
        correct_ans = q["answer"]
        if user_ans == correct_ans:
            score += 1
        
        # Question display
        st.markdown(f"**{texts['question_prefix']}{i+1}: {q['question']}**")
        
        # Answers with color coding
        if user_ans == correct_ans:
            st.markdown(f"- {texts['your_answer']} :green[{user_ans}] ✓")
        else:
            st.markdown(f"- {texts['your_answer']} :red[{user_ans}] ✗")
        st.markdown(f"- {texts['correct_answer']} :green[{correct_ans}]")
        st.markdown("---")

    # Score display
    score_text = texts["score"].format(score=score, total=total)
    st.success(score_text)
    
    # Reset button
    if language == "English":
        reset_text = "Start New Quiz"
    else:
        reset_text = "새 퀴즈 시작"
    
    if st.button(reset_text):
        st.session_state.mcqs = []
        st.session_state.current_q = 0
        st.session_state.user_answers = {}
        st.session_state.quiz_finished = False
        st.rerun()