classify_character_inquiry.py

"""
classify_character_inquiry.py
Detect when the user is asking about other available characters.

LLM-only — no keyword fast-path (keywords caused too many false positives).

Returns:
    {
        "is_character_inquiry": bool,
        "inquiry_type": "style" | "pragmatic" | "general" | None
            style     — user wants someone with a different conversational style
                        ("someone more direct", "less patient", "more sarcastic")
            pragmatic — user wants a character better suited to structured guidance
                        ("someone who explains step by step", "someone more patient")
            general   — generic "who else is there" / "what other characters"
    }
"""
import json
import re
from typing import Dict, Any, List

from llm_client import client
from config import OPENAI_CLASSIFIER_MODEL


def _parse_json(raw: str, fallback: Any) -> Any:
    m = re.search(r"\{.*\}", raw, re.S)
    if m:
        try:
            return json.loads(m.group(0))
        except json.JSONDecodeError:
            pass
    try:
        return json.loads(raw)
    except json.JSONDecodeError:
        return fallback


def classify_character_inquiry(
    user_msg: str,
    history: List[Dict[str, str]],
) -> Dict[str, Any]:
    """
    Detect whether the user is asking about other available characters.
    Pure LLM classification — no keyword matching.
    """
    _NULL = {"is_character_inquiry": False, "inquiry_type": None, "target_character": None}

    # Skip very short messages — they are almost never character inquiries
    if len(user_msg.split()) <= 3:
        return _NULL

    system = (
        "The user is talking with a philosophical chatbot that has multiple characters "
        "(Socrates, Diogenes, Nietzsche, Camus, Schopenhauer). "
        "Detect whether the user's message is asking about other available characters "
        "or requesting a switch to a different one.\n\n"
        "Return ONLY valid JSON:\n"
        '{ "is_character_inquiry": true/false, "inquiry_type": "style" | "pragmatic" | "general" | null, '
        '"target_character": "socrates" | "diogenes" | "nietzsche" | "camus" | "schopenhauer" | null }\n\n'
        "is_character_inquiry = true ONLY when the user:\n"
        "  - Explicitly asks if there is SOMEONE ELSE or ANOTHER CHARACTER available\n"
        "  - Asks for a character with a different personality or style\n"
        "  - Asks who else they could speak to\n"
        "  - Requests to switch to a different character\n\n"
        "is_character_inquiry = false when the user:\n"
        "  - Is asking the CURRENT character to explain something better or in more detail\n"
        "  - Is asking a follow-up question on the current topic\n"
        "  - Is asking for step-by-step guidance from the current character\n"
        "  - Uses phrases like 'walk me through', 'explain more', 'be more detailed'\n"
        "    (these are requests TO the current character, not requests FOR a new one)\n\n"
        "inquiry_type (only when is_character_inquiry = true):\n"
        "  style     — user wants a different conversational style (more direct, gentler, etc.)\n"
        "  pragmatic — user wants someone better at structured, step-by-step guidance\n"
        "  general   — user just wants to know who else is available\n"
        "  null      — not a character inquiry\n\n"
        "target_character = the specific character the user wants to be CONNECTED TO / TALK TO / "
        "SWITCHED TO by name (e.g. 'can you call Nietzsche?', 'let me talk to Camus', "
        "'bring Diogenes', 'pass me to him' referring to a named character). "
        "Set it ONLY when the user wants to actually switch to or speak with that specific person. "
        "Set it to null if the user merely asks what another character THINKS about the topic "
        "(that is a comparison question, not a switch), or names no one specific. "
        "When target_character is set, is_character_inquiry must also be true.\n\n"
        "When in doubt, return false — do not fire on ambiguous phrasing."
    )

    recent = history[-4:] if history else []
    msgs = [{"role": "system", "content": system}]
    for h in recent:
        role = h.get("role", "user")
        if role not in ("user", "assistant"):
            role = "user"
        msgs.append({"role": role, "content": h.get("content", "")})
    msgs.append({"role": "user", "content": user_msg})

    try:
        resp = client.chat.completions.create(
            model=OPENAI_CLASSIFIER_MODEL,
            messages=msgs,
            temperature=0.1,
        )
        raw = (resp.choices[0].message.content or "").strip()
        parsed = _parse_json(raw, _NULL)
        _KNOWN = {"socrates", "diogenes", "nietzsche", "camus", "schopenhauer"}
        _target = (parsed.get("target_character") or "").lower().strip()
        if _target not in _KNOWN:
            _target = None
        return {
            "is_character_inquiry": bool(parsed.get("is_character_inquiry", False)) or bool(_target),
            "inquiry_type": parsed.get("inquiry_type") or None,
            "target_character": _target,
        }
    except Exception:
        return _NULL