proj/backend/app/services/evaluation_service.py

import asyncio
import json
import re
from typing import List, Dict, Any
from app.utils.groq_client import get_groq_completion
from app.models.schemas import (
    Candidate, NormalizedCandidate, RerankResult, 
    DeepReview, FinalShortlist, FinalRank, EvaluationResponse
)
from app.services.matching_service import match_service
from app.prompts.templates import (
    STAGE1_NORMALIZATION_PROMPT,
    STAGE3_RERANK_PROMPT,
    STAGE4_DEEP_REVIEW_PROMPT,
    STAGE5_FINAL_SELECTION_PROMPT
)

# Concurrency Throttling
sem = asyncio.Semaphore(3)

async def get_completion_with_sem(messages):
    async with sem:
        return await get_groq_completion(messages)

async def normalize_candidate(jd: str, candidate: Candidate) -> NormalizedCandidate:
    candidate_raw = candidate.model_dump_json()
    resp = await get_completion_with_sem([
        {"role": "system", "content": "You are a professional data normalizer. Output JSON ONLY."},
        {"role": "user", "content": STAGE1_NORMALIZATION_PROMPT.format(jd=jd, candidate_raw=candidate_raw)}
    ])
    try:
        match = re.search(r'\{.*\}', resp, re.DOTALL)
        data = json.loads(match.group() if match else resp)
        return NormalizedCandidate(**data)
    except Exception as e:
        print(f"Failed to normalize {candidate.name}: {e}")
        # Return a fallback object
        return NormalizedCandidate(
            candidate_id=candidate.id, name=candidate.name, normalized_title="Unknown",
            experience_years=0, primary_skills=[], secondary_skills=[],
            backend_score=0, frontend_score=0, cloud_score=0, database_score=0,
            notice_period_days=0, location="Unknown", employment_status="Unknown",
            salary_expectation="Unknown", flags=["Parsing Error"]
        )

async def rerank_candidate(jd: str, normalized: NormalizedCandidate) -> RerankResult:
    resp = await get_completion_with_sem([
        {"role": "system", "content": "You are a recruitment scoring engine. Output JSON ONLY."},
        {"role": "user", "content": STAGE3_RERANK_PROMPT.format(jd=jd, normalized_candidate=normalized.model_dump_json())}
    ])
    try:
        match = re.search(r'\{.*\}', resp, re.DOTALL)
        data = json.loads(match.group() if match else resp)
        return RerankResult(**data)
    except:
        return RerankResult(candidate_id=normalized.candidate_id, scores={}, final_score=0, decision="reject")

async def review_candidate(jd: str, candidate_data: str, score: float, cand_id: str) -> DeepReview:
    resp = await get_completion_with_sem([
        {"role": "system", "content": "You are a senior hiring evaluator. Output JSON ONLY."},
        {"role": "user", "content": STAGE4_DEEP_REVIEW_PROMPT.format(jd=jd, candidate_data=candidate_data, score=score)}
    ])
    try:
        match = re.search(r'\{.*\}', resp, re.DOTALL)
        data = json.loads(match.group() if match else resp)
        data["candidate_id"] = cand_id
        return DeepReview(**data)
    except:
        return DeepReview(candidate_id=cand_id, verdict="reject", why="Error in evaluation", strengths=[], risks=[], hidden_signal="", confidence=0)

async def perform_hybrid_evaluation(jd: str, candidates: List[Candidate]) -> EvaluationResponse:
    # 1. Normalization (Stage 1) - All candidates
    normalization_tasks = [normalize_candidate(jd, c) for c in candidates]
    normalized_candidates = await asyncio.gather(*normalization_tasks)
    
    # Map for easy lookup
    normalized_map = {n.candidate_id: n for n in normalized_candidates}
    candidate_map = {c.id: c for c in candidates}
    
    # 2. Embedding Matching (Stage 2) - Retrieves Top 20
    # We pass the normalized summary/skills for better matching
    top_20 = await match_service.get_top_candidates(jd, candidates)
    
    # 3. Deterministic Reranking (Stage 3) - Top 20 -> Top 10
    rerank_tasks = [rerank_candidate(jd, normalized_map[c.id]) for c in top_20]
    rerank_results = await asyncio.gather(*rerank_tasks)
    rerank_results.sort(key=lambda x: x.final_score, reverse=True)
    top_10_results = rerank_results[:10]
    
    # 4. LLM Deep Review (Stage 4) - Top 5 Only
    top_5_for_review = top_10_results[:5]
    review_tasks = [
        review_candidate(
            jd, 
            candidate_map[r.candidate_id].model_dump_json(), 
            r.final_score, 
            r.candidate_id
        ) for r in top_5_for_review
    ]
    review_results = await asyncio.gather(*review_tasks)
    review_map = {rev.candidate_id: rev for rev in review_results}
    
    # 5. Final Selection (Stage 5)
    reviews_json = json.dumps([r.model_dump() for r in review_results])
    final_resp = await get_completion_with_sem([
        {"role": "system", "content": "You are the final hiring decision officer. Output JSON ONLY."},
        {"role": "user", "content": STAGE5_FINAL_SELECTION_PROMPT.format(all_top_5_results=reviews_json)}
    ])
    
    try:
        match = re.search(r'\{.*\}', final_resp, re.DOTALL)
        final_data = json.loads(match.group() if match else final_resp)
        shortlist = FinalShortlist(**final_data)
    except:
        # Fallback ranking if synthesis fails
        shortlist = FinalShortlist(final_ranking=[
            FinalRank(rank=i+1, candidate_id=r.candidate_id, name=candidate_map[r.candidate_id].name, decision=r.decision, reason="Automatic ranking")
            for i, r in enumerate(top_5_for_review)
        ])

    return EvaluationResponse(
        shortlist=shortlist.final_ranking,
        details={rev.candidate_id: rev.model_dump() for rev in review_results}
    )