import os
import gradio as gr
from llama_index.readers.file import PDFReader
from llama_index.core import VectorStoreIndex
from llama_index.llms.openai import OpenAI as LlamaOpenAI
from openai import OpenAI  # New SDK
from gtts import gTTS
import random
import tempfile
import speech_recognition as sr

# 🔐 Set API key directly (use HF secret or hardcode for testing)
OPENAI_API_KEY = os.environ.get('HF_OPENAI_API_KEY', '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')  # Replace or use HF secret
client = OpenAI(api_key=OPENAI_API_KEY)

# Global state
query_engine = None
resume_summary = ""
questions = []
asked_questions = []

# Step 1: Load and index resume
def load_resume(file):
    global query_engine, resume_summary, questions, asked_questions

    reader = PDFReader()
    documents = reader.load_data(file=file.name)
    index = VectorStoreIndex.from_documents(documents)
    query_engine = index.as_query_engine()

    resume_summary = query_engine.query("Summarize this resume in detail.")
    prompt = f"Generate 5 personalized interview-style questions based on this resume:\n{resume_summary}"
    questions = [str(query_engine.query(prompt + f"\nQuestion {i+1}")) for i in range(5)]
    asked_questions = []

    return f"✅ Resume analyzed!\n\n📄 Summary:\n{resume_summary}"

# Step 2: Ask next question (TTS with gTTS)
def ask_next_question():
    global questions, asked_questions

    if not questions:
        return "❌ Resume not loaded yet.", None

    remaining = [q for q in questions if q not in asked_questions]
    if not remaining:
        return "✅ All questions have been asked.", None

    next_q = random.choice(remaining)
    asked_questions.append(next_q)

    # Convert text to audio using gTTS
    tts = gTTS(next_q)
    temp_audio = tempfile.NamedTemporaryFile(delete=False, suffix=".mp3")
    tts.save(temp_audio.name)

    return next_q, temp_audio.name

# Step 3: Speech to Text
def speech_to_text(audio_path):
    recognizer = sr.Recognizer()
    with sr.AudioFile(audio_path) as source:
        audio_data = recognizer.record(source)
    try:
        return recognizer.recognize_google(audio_data)
    except sr.UnknownValueError:
        return "⚠️ Could not understand the voice input."
    except sr.RequestError:
        return "⚠️ API error with speech recognition."

# Step 4: Evaluate answer using OpenAI SDK v1.0+
def evaluate_answer(audio_path):
    if not audio_path:
        return "⚠️ Please record an answer.", "❌ No score"

    transcript = speech_to_text(audio_path)
    if transcript.startswith("⚠️"):
        return transcript, "❌"

    rating_prompt = f"""
You are an interview evaluator. Rate the following candidate's answer from 1 to 5 based on:
- Relevance
- Clarity
- Completeness

Answer: "{transcript}"
Only return: Score: <number>
"""

    try:
        response = client.chat.completions.create(
            model="gpt-3.5-turbo",
            messages=[{"role": "user", "content": rating_prompt}],
            max_tokens=10,
            temperature=0.2,
        )
        score = response.choices[0].message.content.strip()
        return f"📝 Transcript: {transcript}", f"🔎 {score}"
    except Exception as e:
        return "❌ Evaluation failed", str(e)

# Gradio Interface
with gr.Blocks() as demo:
    gr.Markdown("# 🎤 Voice Resume Interview Bot\nUpload your resume → get questions in voice → answer via mic → get score.")

    with gr.Row():
        resume_file = gr.File(label="📄 Upload Resume", file_types=[".pdf"])
        resume_summary_box = gr.Textbox(label="Resume Summary", lines=6)
    gr.Button("Analyze Resume").click(fn=load_resume, inputs=resume_file, outputs=resume_summary_box)

    gr.Markdown("## 🔊 Ask a Resume-Based Question")
    question_text = gr.Textbox(label="Question", lines=2)
    question_audio = gr.Audio(label="Voice", type="filepath", autoplay=True)
    gr.Button("Ask Question").click(fn=ask_next_question, outputs=[question_text, question_audio])

    gr.Markdown("## 🎙️ Speak Your Answer")
    user_audio = gr.Audio(label="Record Your Answer", type="filepath")
    transcript_output = gr.Textbox(label="Transcript")
    score_output = gr.Textbox(label="LLM Score")
    gr.Button("Submit Answer").click(fn=evaluate_answer, inputs=user_audio, outputs=[transcript_output, score_output])

demo.launch()