import time
import logging
from dataclasses import dataclass, field
from datetime import datetime
from prometheus_client import Counter, Histogram, Gauge

# Prometheus metrics
ai_requests_total = Counter('ai_requests_total', 'Total AI requests', ['model', 'status'])
ai_latency_seconds = Histogram('ai_latency_seconds', 'AI request latency', ['model'])
ai_cost_dollars = Counter('ai_cost_dollars', 'Total AI cost in dollars', ['model'])
ai_quality_score = Gauge('ai_quality_score', 'Moving average quality score')

@dataclass
class AIRequestMetrics:
    request_id: str
    model: str
    input_tokens: int
    output_tokens: int
    latency_ms: float
    cost_usd: float
    quality_score: float = None
    error: str = None
    timestamp: str = field(default_factory=lambda: datetime.now().isoformat())

def monitored_llm_call(messages: list, model: str = "gpt-4o") -> tuple[str, AIRequestMetrics]:
    """Wrap LLM calls with comprehensive monitoring."""
    start_time = time.time()
    request_id = f"req_{int(time.time() * 1000)}"

    try:
        client = openai.OpenAI()
        response = client.chat.completions.create(
            model=model,
            messages=messages,
            temperature=0
        )

        latency = (time.time() - start_time) * 1000
        cost = calculate_cost(model, response.usage.prompt_tokens, response.usage.completion_tokens)

        metrics = AIRequestMetrics(
            request_id=request_id,
            model=model,
            input_tokens=response.usage.prompt_tokens,
            output_tokens=response.usage.completion_tokens,
            latency_ms=latency,
            cost_usd=cost
        )

        # Update Prometheus metrics
        ai_requests_total.labels(model=model, status="success").inc()
        ai_latency_seconds.labels(model=model).observe(latency / 1000)
        ai_cost_dollars.labels(model=model).inc(cost)

        return response.choices[0].message.content, metrics

    except Exception as e:
        latency = (time.time() - start_time) * 1000
        metrics = AIRequestMetrics(
            request_id=request_id, model=model,
            input_tokens=0, output_tokens=0,
            latency_ms=latency, cost_usd=0, error=str(e)
        )
        ai_requests_total.labels(model=model, status="error").inc()
        raise
AI in Production — MLOps, Monitoring, Evaluation

AI in Production — MLOps, Monitoring, Evaluation

AI in Production — MLOps, Monitoring, Evaluation
