src/evaluator.py

# File: src/evaluator.py
# Purpose: Score candidate responses using Groq and produce a hiring report

import json
import re
from typing import List, Dict, Any

from groq import Groq

from config import GROQ_API_KEY, GROQ_LLM_MODEL, SCORE_WEIGHTS

_groq = Groq(api_key=GROQ_API_KEY)

EVALUATION_PROMPT = """You are an expert technical recruiter evaluating a candidate's interview.
Score the candidate on these three dimensions (0-10 scale):
1. communication_score: Clarity, structure, and articulation of responses.
2. technical_score: Depth, accuracy, and relevance of technical knowledge.
3. confidence_score: Certainty, specificity, and ownership shown in answers.

Also extract:
- skills: JSON array of technical skills demonstrated.
- strengths: 2-3 bullet points.
- weaknesses: 1-2 bullet points.
- recommendation: One of ["Proceed to Technical Round", "Proceed with Caution", "Do Not Proceed"].
- summary: One paragraph summary for the recruiter.

Respond ONLY with a valid JSON object. No markdown, no extra text.
Schema:
{
  "communication_score": <float>,
  "technical_score": <float>,
  "confidence_score": <float>,
  "skills": [<string>],
  "strengths": [<string>],
  "weaknesses": [<string>],
  "recommendation": <string>,
  "summary": <string>
}
"""


def evaluate_candidate(candidate_name: str, transcript: List[Dict[str, str]]) -> Dict[str, Any]:
    transcript_text = "\n".join(f"{t['speaker']}: {t['text']}" for t in transcript)
    response = _groq.chat.completions.create(
        model=GROQ_LLM_MODEL,
        messages=[
            {"role": "system", "content": EVALUATION_PROMPT},
            {"role": "user", "content": f"Candidate: {candidate_name}\n\nTranscript:\n{transcript_text}"},
        ],
        temperature=0,
        max_tokens=800,
    )
    raw = response.choices[0].message.content.strip()
    raw = re.sub(r"```json|```", "", raw).strip()
    result = json.loads(raw)
    overall = (
        result["communication_score"] * SCORE_WEIGHTS["communication"]
        + result["technical_score"] * SCORE_WEIGHTS["technical_skills"]
        + result["confidence_score"] * SCORE_WEIGHTS["confidence"]
    )
    result["overall_score"] = round(overall, 2)
    return result


def format_report(candidate_name: str, phone: str, evaluation: Dict[str, Any]) -> str:
    skills_str = "\n".join(f"  - {s}" for s in evaluation.get("skills", []))
    strengths_str = "\n".join(f"  - {s}" for s in evaluation.get("strengths", []))
    weaknesses_str = "\n".join(f"  - {w}" for w in evaluation.get("weaknesses", []))
    return f"""
AI Interview Screening Report
==============================
Candidate : {candidate_name}
Phone     : {phone}

Scores
------
Communication  : {evaluation['communication_score']:.1f} / 10
Technical      : {evaluation['technical_score']:.1f} / 10
Confidence     : {evaluation['confidence_score']:.1f} / 10
Overall        : {evaluation['overall_score']:.1f} / 10

Skills Identified
-----------------
{skills_str}

Strengths
---------
{strengths_str}

Areas to Probe Further
-----------------------
{weaknesses_str}

Summary
-------
{evaluation.get('summary', '')}

Recommendation
--------------
{evaluation.get('recommendation', 'N/A')}
==============================
""".strip()