src/api/chat_routes.py

"""
Chat routes — Document-specific Q&A powered by local Qwen model.
"""

from typing import List, Optional

from fastapi import APIRouter, Depends, HTTPException
from pydantic import BaseModel, Field

from src.auth.dependencies import get_current_user
from src.db.models import User
from src.summarization.note_generator import NoteGenerator
from src.utils.logger import setup_logger
from src.utils.model_loader import INFERENCE_MODE, get_groq_client

logger = setup_logger(__name__)
router = APIRouter(tags=["Chat"])


# ─── Schemas ──────────────────────────────────────────────────────────

class ChatMessage(BaseModel):
    role: str = Field(..., description="Either 'user' or 'assistant'")
    content: str = Field(..., description="Message content")


class ChatRequest(BaseModel):
    note_content: str = Field(
        ...,
        description="The full text of the note to use as context",
    )
    question: str = Field(
        ...,
        min_length=1,
        description="The user's question about the note",
    )
    history: Optional[List[ChatMessage]] = Field(
        default=None,
        description="Previous conversation turns for multi-turn context",
    )


class ChatResponse(BaseModel):
    answer: str = Field(..., description="AI-generated answer")


# ─── Endpoint ─────────────────────────────────────────────────────────

@router.post("/chat/note", response_model=ChatResponse)
async def chat_with_note(
    request: ChatRequest,
    current_user: User = Depends(get_current_user),
):
    """Ask a question about a specific note. Answers are grounded in the note content."""
    logger.info(
        "Chat request from user %s — question length: %d, context length: %d",
        current_user.id,
        len(request.question),
        len(request.note_content),
    )

    if not request.note_content.strip():
        raise HTTPException(status_code=400, detail="Note content cannot be empty.")

    history_dicts = None
    if request.history:
        history_dicts = [msg.model_dump() for msg in request.history]

    if INFERENCE_MODE == "groq":
        logger.info("🟢 Chat request routed directly to Groq API (llama-3.3-70b-versatile)...")
        messages = [
            {
                "role": "system",
                "content": (
                    "You are a helpful study assistant. "
                    "Answer the user's question based ONLY on the note content provided below. "
                    "If the answer is not in the note, say so honestly. "
                    "Reply in the same language the user uses.\n\n"
                    f"--- NOTE CONTENT ---\n{request.note_content[:4000]}\n--- END NOTE ---"
                ),
            },
        ]
        if history_dicts:
            for msg in history_dicts[-6:]:
                role = msg.get("role", "user")
                content = msg.get("content", "")
                if role in ("user", "assistant") and content:
                    messages.append({"role": role, "content": content})

        messages.append({"role": "user", "content": request.question})

        groq_client = get_groq_client()
        try:
            chat_completion = groq_client.chat.completions.create(
                model="llama-3.3-70b-versatile",
                messages=messages,
                max_tokens=300,
                temperature=0.0,
            )
            answer = chat_completion.choices[0].message.content or ""
            answer = answer.strip()
            if not answer or len(answer) < 3:
                answer = "عذرًا، لم أتمكن من توليد إجابة. يرجى إعادة صياغة السؤال."
        except Exception as e:
            logger.error("❌ Groq chat error: %s", e, exc_info=True)
            answer = "حدث خطأ أثناء معالجة سؤالك. يرجى المحاولة مرة أخرى."
    else:
        logger.info("🤖 Chat request routed to local Qwen pipeline...")
        note_gen = NoteGenerator()
        answer = note_gen.chat_with_note(
            note_content=request.note_content,
            question=request.question,
            history=history_dicts,
        )

    return ChatResponse(answer=answer)