"""FluentGerman.ai — LLM service (provider-agnostic via LiteLLM)."""

import logging
from collections.abc import AsyncGenerator

import litellm

from app.config import get_settings

logger = logging.getLogger("fluentgerman.llm")


def _resolve_model(model: str, provider: str) -> str:
    """Ensure Gemini models have the 'gemini/' prefix for Google AI Studio."""
    if provider == "gemini" and not model.startswith("gemini/"):
        resolved = f"gemini/{model}"
        logger.info("Auto-prefixed model: %s → %s (Google AI Studio)", model, resolved)
        return resolved
    return model


async def chat_stream(
    messages: list[dict[str, str]],
    model: str | None = None,
) -> AsyncGenerator[str, None]:
    """Stream chat completion tokens from the configured LLM provider."""
    settings = get_settings()
    model = _resolve_model(model or settings.llm_model, settings.llm_provider)

    logger.info("LLM request: model=%s messages=%d", model, len(messages))

    response = await litellm.acompletion(
        model=model,
        messages=messages,
        api_key=settings.llm_api_key,
        stream=True,
    )

    async for chunk in response:
        delta = chunk.choices[0].delta
        if delta.content:
            yield delta.content


async def summarize_instruction(raw_text: str) -> str:
    """Ask the LLM to distill raw voice transcript into a structured instruction."""
    settings = get_settings()
    model = _resolve_model(settings.llm_model, settings.llm_provider)

    meta_prompt = (
        "You are an expert assistant for a language teacher. "
        "The following text is a raw transcript of the teacher describing a learning instruction, "
        "homework, or teaching method. Distill it into a clear, concise, structured instruction "
        "that can be used as a system prompt for a language-learning LLM assistant. "
        "Output ONLY the instruction text, no preamble.\n\n"
        f"Transcript:\n{raw_text}"
    )

    response = await litellm.acompletion(
        model=model,
        messages=[{"role": "user", "content": meta_prompt}],
        api_key=settings.llm_api_key,
    )

    return response.choices[0].message.content.strip()