import gradio as gr
import requests
import os
import random

# =====================================
# CONFIG
# =====================================

HF_TOKEN = os.getenv("HF_TOKEN")

WHISPER_API = "https://api-inference.huggingface.co/models/openai/whisper-base"
LLM_API = "https://api-inference.huggingface.co/models/mistralai/Mistral-7B-Instruct-v0.2"

headers = {
    "Authorization": f"Bearer {HF_TOKEN}"
}

# =====================================
# Question Bank
# =====================================

questions = {
    "Easy": [
        "What is Machine Learning?",
        "Explain supervised learning.",
        "What is overfitting?"
    ],
    "Medium": [
        "Explain bias vs variance tradeoff.",
        "What is gradient descent?",
        "Difference between CNN and RNN?"
    ],
    "Hard": [
        "Explain backpropagation.",
        "What is attention mechanism?",
        "Explain transformer architecture."
    ]
}

# =====================================
# Generate Question
# =====================================

def start_interview(level):
    return random.choice(questions[level])

# =====================================
# Speech-to-Text using API
# =====================================

def speech_to_text(audio_path):
    try:
        with open(audio_path, "rb") as f:
            audio_bytes = f.read()

        response = requests.post(
            WHISPER_API,
            headers=headers,
            data=audio_bytes
        )

        if response.status_code == 200:
            return response.json().get("text", "Could not transcribe audio.")
        else:
            return "Speech recognition API error."

    except Exception as e:
        return f"Speech recognition failed: {str(e)}"


# =====================================
# LLM Evaluation
# =====================================

def evaluate_with_llm(question, answer):

    prompt = f"""
You are a strict technical interviewer.

Question:
{question}

Candidate Answer:
{answer}

Evaluate and give:

1. Technical Accuracy (0-10)
2. Clarity (0-10)
3. Depth (0-10)
4. Overall Score (0-10)
5. Improvement Suggestions

Be concise and structured.
"""

    payload = {
        "inputs": prompt,
        "parameters": {
            "max_new_tokens": 300,
            "temperature": 0.7
        }
    }

    try:
        response = requests.post(
            LLM_API,
            headers=headers,
            json=payload
        )

        if response.status_code == 200:
            return response.json()[0]["generated_text"]
        else:
            return "LLM evaluation API error."

    except Exception as e:
        return f"Evaluation failed: {str(e)}"


# =====================================
# Main Evaluation Function
# =====================================

def evaluate_answer(audio, question):

    if audio is None:
        return "Please record your answer."

    if question.strip() == "":
        return "Please click 'Start Interview' first."

    transcribed_text = speech_to_text(audio)

    feedback = evaluate_with_llm(question, transcribed_text)

    return f"""
📝 Transcribed Answer:
{transcribed_text}

📊 Evaluation:
{feedback}
"""


# =====================================
# UI
# =====================================

with gr.Blocks() as demo:

    gr.Markdown("# 🎤 Smart Interview Simulator")
    gr.Markdown("Voice-Based AI Mock Interview System")

    level_dropdown = gr.Dropdown(
        ["Easy", "Medium", "Hard"],
        value="Medium",
        label="Select Difficulty"
    )

    question_output = gr.Textbox(label="Interview Question")

    start_button = gr.Button("Start Interview")
    start_button.click(
        start_interview,
        inputs=level_dropdown,
        outputs=question_output
    )

    audio_input = gr.Audio(
        type="filepath",
        label="Record Your Answer"
    )

    submit_button = gr.Button("Submit Answer")

    result_output = gr.Textbox(
        label="Evaluation Feedback",
        lines=15
    )

    submit_button.click(
        evaluate_answer,
        inputs=[audio_input, question_output],
        outputs=result_output
    )

demo.launch()