import gradio as gr
import requests
import os
import re

# We use Mistral because DialoGPT is for chatting, while Mistral follows instructions
API_URL = "https://api-inference.huggingface.co/models/mistralai/Mistral-7B-Instruct-v0.3"
HF_TOKEN = os.getenv("HF_TOKEN")

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

def query_llm(prompt):
    """Helper function to send prompts to the Hugging Face API."""
    payload = {
        "inputs": f"<s>[INST] {prompt} [/INST]",
        "parameters": {"max_new_tokens": 250, "temperature": 0.7}
    }
    try:
        response = requests.post(API_URL, headers=headers, json=payload)
        result = response.json()
        
        if isinstance(result, list) and "generated_text" in result[0]:
            # Clean the output to remove the prompt markers
            full_text = result[0]["generated_text"]
            return full_text.split("[/INST]")[-1].strip()
        
        return "The AI is still warming up. Please wait 30 seconds and try again."
    except Exception as e:
        return f"Error: Could not connect to AI (Check your HF_TOKEN). {str(e)}"

# ----------------------------
# AI Logic Functions
# ----------------------------

def generate_question(role, difficulty, resume_text):
    if not role:
        return "Please enter a Job Role first!"
        
    prompt = f"Act as a professional recruiter. Generate one {difficulty} level technical interview question for a {role} role."
    if resume_text.strip():
        prompt += f" Based on this resume context: {resume_text}"
    
    return query_llm(prompt)

def evaluate_answer(question, answer):
    if not answer or len(answer) < 5:
        return "Please provide a more detailed answer for evaluation.", "N/A"
        
    prompt = f"""
    Interviewer Question: {question}
    Candidate Answer: {answer}
    
    Task: Critically evaluate this answer. 
    1. Give constructive feedback.
    2. Provide a score out of 10.
    Format your response with the score clearly at the end as 'Final Score: X/10'.
    """
    
    feedback = query_llm(prompt)
    
    # Extract score using regex (looks for X/10)
    score_match = re.search(r"(\d+/10)", feedback)
    score = score_match.group(1) if score_match else "Score not generated"
    
    return feedback, score

# ----------------------------
# Gradio UI Design
# ----------------------------

with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown("# 🤖 Smart Interview Simulator")
    gr.Markdown("Practice your interview skills with AI-generated questions and real-time feedback.")

    with gr.Row():
        with gr.Column():
            role_input = gr.Textbox(label="Target Job Role", placeholder="e.g., Python Developer")
            diff_input = gr.Dropdown(["Easy", "Medium", "Hard"], label="Level", value="Medium")
            resume_input = gr.Textbox(label="Resume Summary (Optional)", lines=3)
            gen_btn = gr.Button("Generate Interview Question", variant="primary")
        
        with gr.Column():
            question_box = gr.Textbox(label="AI Question", lines=5, interactive=False)

    gr.Markdown("---")

    with gr.Row():
        with gr.Column():
            gr.Markdown("### Your Response")
            # Note: For voice input to work, Gradio handles the file, 
            # but you would need a transcription model (like Whisper) to convert audio to text.
            # For now, we will focus on the text input.
            ans_input = gr.Textbox(label="Type your answer here", lines=5)
            eval_btn = gr.Button("Submit for Evaluation", variant="secondary")
            
        with gr.Column():
            gr.Markdown("### Results")
            feedback_box = gr.Textbox(label="AI Feedback", lines=5)
            score_box = gr.Label(label="Final Score")

    # Button actions
    gen_btn.click(
        fn=generate_question, 
        inputs=[role_input, diff_input, resume_input], 
        outputs=question_box
    )
    
    eval_btn.click(
        fn=evaluate_answer, 
        inputs=[question_box, ans_input], 
        outputs=[feedback_box, score_box]
    )

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