src/agents/reflector.py

from __future__ import annotations

import asyncio
import json
import logging
from typing import Any

from src.agents.prompts import REFLECTOR_SYSTEM_PROMPT
from src.core.config import get_llm_client, get_settings
from src.core.models import MatchResult, ParsedQuery

logger = logging.getLogger(__name__)


class ReflectorAgent:
    def __init__(self) -> None:
        self._client = None
        settings = get_settings()
        self.model = settings.openai_model

    @property
    def client(self):
        if self._client is None:
            try:
                self._client = get_llm_client()
            except Exception:
                logger.warning("LLM client unavailable for reflector")
                self._client = None
        return self._client

    async def reflect(self, query: ParsedQuery, results: list[MatchResult]) -> dict[str, Any]:
        try:
            if self.client is None:
                raise RuntimeError("LLM client unavailable")

            from langchain_core.messages import HumanMessage, SystemMessage

            candidates_text = json.dumps(
                [
                    {
                        "profile_id": r.profile_id,
                        "name": r.name,
                        "title": r.current_title,
                        "company": r.current_company,
                        "overall_score": r.scores.overall,
                        "confidence": r.scores.confidence,
                        "matched_skills": r.matched_skills[:10],
                        "missing_skills": r.missing_skills[:10],
                    }
                    for r in results[:20]
                ],
                indent=2,
            )

            messages = [
                SystemMessage(content=REFLECTOR_SYSTEM_PROMPT),
                HumanMessage(content=f"Candidates:\n{candidates_text}"),
            ]
            response = await asyncio.wait_for(
                self.client.ainvoke(messages),
                timeout=30.0,
            )
            content = response.content if hasattr(response, "content") else str(response)
            if not content or not content.strip():
                logger.warning("Reflector LLM returned empty content, using fallback")
                evaluations = self._fallback_evaluate(results)
            else:
                evaluations = json.loads(content)

        except TimeoutError:
            logger.warning("Reflector LLM timed out, using fallback")
            evaluations = self._fallback_evaluate(results)
        except Exception as e:
            logger.warning(f"Reflector LLM failed: {type(e).__name__}: {e}, using fallback")
            evaluations = self._fallback_evaluate(results)

        should_replan = self._should_replan(evaluations)
        good_matches = {"strong_match", "good_match"}
        good_count = sum(
            1 for ev in (evaluations if isinstance(evaluations, list) else [])
            if isinstance(ev, dict) and ev.get("overall_assessment") in good_matches
        )

        return {
            "evaluations": evaluations if isinstance(evaluations, list) else [],
            "good_match_count": good_count,
            "should_replan": should_replan,
            "feedback": self._generate_feedback(evaluations, results),
        }

    def _fallback_evaluate(self, results: list[MatchResult]) -> list[dict]:
        evaluations: list[dict] = []
        for r in results[:20]:
            if r.scores.overall >= 0.8:
                assessment = "strong_match"
            elif r.scores.overall >= 0.6:
                assessment = "good_match"
            elif r.scores.overall >= 0.4:
                assessment = "potential_match"
            else:
                assessment = "weak_match"

            evaluations.append({
                "profile_id": r.profile_id,
                "overall_assessment": assessment,
                "key_strengths": r.matched_skills[:5],
                "key_gaps": r.missing_skills[:5],
                "concerns": [],
                "should_keep": assessment in ("strong_match", "good_match", "potential_match"),
            })
        return evaluations

    def _should_replan(  # noqa: E501
        self, evaluations: list[dict[str, Any]] | dict[str, Any], threshold: int = 3
    ) -> bool:
        if isinstance(evaluations, list):
            good_matches = {"strong_match", "good_match"}
            good = sum(
                1 for ev in evaluations
                if isinstance(ev, dict) and ev.get("overall_assessment") in good_matches
            )
            return good < threshold
        return False

    def _generate_feedback(  # noqa: E501
        self, evaluations: list[dict[str, Any]] | dict[str, Any], results: list[MatchResult]
    ) -> str:
        if isinstance(evaluations, list):
            weak = [
                ev for ev in evaluations
                if isinstance(ev, dict) and ev.get("overall_assessment") == "weak_match"
            ]
            if weak:
                return f"{len(weak)} weak matches found. Consider broadening criteria."
        if not results:
            return "No results found. Query may be too restrictive."
        return "Results look reasonable."