import os
import time
import random
import sys
import subprocess
from typing import List
from fastapi import FastAPI
from pydantic import BaseModel

# SETUP
try:
    import ollama
    from sentence_transformers import SentenceTransformer
    from sklearn.metrics.pairwise import cosine_similarity
except ImportError:
    subprocess.check_call([sys.executable, "-m", "pip", "install", "ollama", "sentence-transformers", "scikit-learn"])
    import ollama
    from sentence_transformers import SentenceTransformer
    from sklearn.metrics.pairwise import cosine_similarity

app = FastAPI(title="AI Interviewer API - Hugging Face", version="2.0")

print("⏳ Loading Judge Model...")
judge_model = SentenceTransformer('paraphrase-multilingual-mpnet-base-v2')
OLLAMA_MODEL = "deepseek-r1:8b"

def ensure_ollama_model():
    print("Checking Ollama service...")
    subprocess.Popen("ollama serve", shell=True)
    time.sleep(5)
    print(f"Pulling model {OLLAMA_MODEL}...")
    subprocess.run(f"ollama pull {OLLAMA_MODEL}", shell=True)
    print("✅ Model ready!")

ensure_ollama_model()

# MODELS
class IntroRequest(BaseModel):
    candidate_intro: str
    track: str
    persona: str

class QuestionRequest(BaseModel):
    track: str
    difficulty: str
    persona: str
    current_q_num: int
    prev_questions: List[str] = []

class AnswerRequest(BaseModel):
    question: str
    ideal_answer: str
    user_answer: str
    time_taken: float
    time_limit: int

class FeedbackRequest(BaseModel):
    persona: str
    score: float
    question: str
    user_answer: str

class SummaryItem(BaseModel):
    score: float
    q: str

class SummaryRequest(BaseModel):
    track: str
    results: List[SummaryItem]

# HELPERS
def get_persona_traits(persona):
    if persona == "Friendly": return "supportive, smiling"
    elif persona == "Aggressive": return "strict, impatient"
    else: return "formal, neutral"

def calculate_score_logic(ref, user_ans):
    if not user_ans.strip(): return 0.0
    emb = judge_model.encode([ref, user_ans])
    return round(float(max(0, cosine_similarity([emb[0]], [emb[1]])[0][0] * 10)), 2)

# ENDPOINTS
@app.get("/")
def home():
    return {"status": "Online on Hugging Face", "model": OLLAMA_MODEL}

@app.post("/analyze_intro")
def analyze_intro(req: IntroRequest):
    traits = get_persona_traits(req.persona)
    prompt = f"Act as {traits} interviewer for {req.track}. Evaluate intro: '{req.candidate_intro}'. Max 2 sentences."
    try:
        res = ollama.chat(model=OLLAMA_MODEL, messages=[{'role': 'user', 'content': prompt}])
        text = res['message']['content'].split("</think>")[-1].strip()
        return {"feedback": text.replace('"', '')}
    except: return {"feedback": "Proceeding."}

@app.post("/generate_question")
def generate_question(req: QuestionRequest):
    base_time = 45
    hist = "\n".join(req.prev_questions[-3:]) if req.prev_questions else "None"
    prompt = f"Ask 1 verbal question about {req.track} ({req.difficulty}). Prev: {hist}. Format: Emotion: [] Question: [] Answer: []"
    
    try:
        res = ollama.chat(model=OLLAMA_MODEL, messages=[{'role': 'user', 'content': prompt}])
        full = res['message']['content'].split("</think>")[-1].strip()
        q, a, e = "Error", "Gen", "Neutral"
        for l in full.split('\n'):
            if "Question:" in l: q = l.split("Question:")[1].strip()
            elif "Answer:" in l: a = l.split("Answer:")[1].strip()
            elif "Emotion:" in l: e = l.split("Emotion:")[1].strip()
        
        if q != "Error": return {"question": q, "ideal_answer": a, "emotion": e, "time_limit": base_time}
    except: pass
    return {"question": f"Explain a core concept in {req.track}.", "ideal_answer": "General", "emotion": "Neutral", "time_limit": 45}

@app.post("/evaluate_answer")
def evaluate_answer(req: AnswerRequest):
    score = calculate_score_logic(req.ideal_answer, req.user_answer)
    final = score
    msg = ""
    if score >= 5:
        if req.time_taken < req.time_limit * 0.7: final += 0.5; msg = "Speed Bonus"
        elif req.time_taken > req.time_limit * 1.5: final -= 0.5; msg = "Late Penalty"
    return {"base_score": score, "final_score": max(0, min(10, final)), "bonus_message": msg}

@app.post("/generate_feedback")
def generate_feedback(req: FeedbackRequest):
    prompt = f"Role: {req.persona}. Q: {req.question}. Ans: {req.user_answer}. Score: {req.score}. 1 sentence feedback."
    try:
        res = ollama.chat(model=OLLAMA_MODEL, messages=[{'role': 'user', 'content': prompt}])
        return {"feedback": res['message']['content'].split("</think>")[-1].strip().replace('"','')}
    except: return {"feedback": "Good effort."}

@app.post("/generate_summary")
def generate_summary(req: SummaryRequest):
    avg = sum(r.score for r in req.results) / len(req.results)
    prompt = f"Summary for {req.track}. Avg: {avg:.1f}. 30 words max."
    try:
        res = ollama.chat(model=OLLAMA_MODEL, messages=[{'role': 'user', 'content': prompt}])
        return {"summary": res['message']['content'].split("</think>")[-1].strip().replace('"','')}
    except: return {"summary": "Good job."}