Spaces:
Sleeping
Sleeping
Download app/services/explainer.py from AnishDev/coderound-task-backend: direct link, hf CLI and curl.
- Browser
- Download file 6.95 kB
-
https://huggingface.co/spaces/AnishDev/coderound-task-backend/resolve/ff7b177dbb49e33cde710ae75dab1dac114fa558/app/services/explainer.py
- Command line
-
hf download hf://spaces/AnishDev/coderound-task-backend@ff7b177dbb49e33cde710ae75dab1dac114fa558/app/services/explainer.py
-
curl -L -o explainer.py https://huggingface.co/spaces/AnishDev/coderound-task-backend/resolve/ff7b177dbb49e33cde710ae75dab1dac114fa558/app/services/explainer.py
6.95 kB
| """ | |
| Lazy explanation generation via Ollama cloud (gpt-oss:120b-cloud). | |
| Called only when a human explicitly requests a candidate detail view. | |
| Result cached in Redis/memory for 48h — subsequent requests return instantly. | |
| Fallback: if Ollama is unavailable, returns a template explanation. | |
| """ | |
| import json | |
| import logging | |
| from ollama import Client | |
| logger = logging.getLogger(__name__) | |
| _ollama_client = None | |
| def _get_client(): | |
| global _ollama_client | |
| if _ollama_client is None: | |
| from app.config import settings | |
| _ollama_client = Client( | |
| host=settings.ollama_host, | |
| headers={"Authorization": f"Bearer {settings.ollama_api_key}"}, | |
| ) | |
| return _ollama_client | |
| def _build_prompt(jd_data: dict, candidate_detail: dict) -> str: | |
| jd_title = jd_data.get("title", "the role") | |
| jd_type = jd_data.get("engineer_type", "") | |
| min_years = jd_data.get("min_years", 0) | |
| req_skills = jd_data.get("required_skills", []) | |
| if isinstance(req_skills, str): | |
| req_skills = json.loads(req_skills) | |
| req_skill_names = [s.get("name", s) if isinstance(s, dict) else s for s in req_skills[:10]] | |
| responsibilities = jd_data.get("responsibilities", []) | |
| if isinstance(responsibilities, str): | |
| responsibilities = json.loads(responsibilities) | |
| name = candidate_detail.get("candidate_name", "Candidate") | |
| position = candidate_detail.get("current_position", "") | |
| years = candidate_detail.get("years_of_experience", 0) | |
| skills_text = candidate_detail.get("parsed_skills", "")[:400] | |
| summary_text = candidate_detail.get("parsed_summary", "")[:300] | |
| work_text = candidate_detail.get("parsed_work_experience_text", "")[:600] | |
| breakdown = candidate_detail.get("score_breakdown", {}) | |
| if isinstance(breakdown, str): | |
| breakdown = json.loads(breakdown) | |
| skill_score = breakdown.get("skill_score", 0) | |
| semantic_score = breakdown.get("semantic_score", 0) | |
| final_score = candidate_detail.get("final_score", 0) | |
| tier = candidate_detail.get("tier", "") | |
| matched = breakdown.get("matched_skills", []) | |
| if isinstance(matched, str): | |
| matched = json.loads(matched) | |
| matched_names = [s.get("skill", s) if isinstance(s, dict) else s for s in matched[:6]] | |
| below = breakdown.get("below_level_skills", []) | |
| if isinstance(below, str): | |
| below = json.loads(below) | |
| below_details = [] | |
| for s in below[:4]: | |
| if isinstance(s, dict): | |
| below_details.append( | |
| f"{s.get('skill','?')}: has level {s.get('candidate_level','?')}, " | |
| f"needs level {s.get('required_level','?')}" | |
| ) | |
| missing = breakdown.get("missing_skills", []) | |
| if isinstance(missing, str): | |
| missing = json.loads(missing) | |
| flags = candidate_detail.get("flags", []) | |
| if isinstance(flags, str): | |
| flags = json.loads(flags) | |
| return f"""You are an expert technical recruiter. Analyze this candidate match and write a concise, specific assessment. | |
| JOB REQUIREMENTS: | |
| - Role: {jd_title} ({jd_type}) | |
| - Minimum Experience: {min_years} years | |
| - Key Required Skills: {', '.join(req_skill_names)} | |
| - Responsibilities (sample): {'; '.join(responsibilities[:3])} | |
| CANDIDATE PROFILE: | |
| - Name: {name} | |
| - Current Role: {position} | |
| - Years of Experience: {years} | |
| - Skills: {skills_text} | |
| - Summary: {summary_text} | |
| - Work History: {work_text} | |
| MATCH ANALYSIS: | |
| - Overall Score: {final_score:.0%} — {tier} | |
| - Skills Match Score: {skill_score:.0%} | |
| - Experience Relevance: {semantic_score:.0%} | |
| - Matched Skills: {', '.join(matched_names) or 'None identified'} | |
| - Below-Level Skills: {'; '.join(below_details) or 'None'} | |
| - Missing Skills: {', '.join(missing[:5]) or 'None'} | |
| - Flags: {', '.join(flags) or 'None'} | |
| Write a structured assessment with EXACTLY these sections: | |
| **Overall Fit** (2 sentences): Reference specific companies and technologies from work history. Be direct and factual. | |
| **Strengths** (2–3 bullets): Each bullet must cite actual data — company names, specific technologies, durations, measurable impacts if available. | |
| **Gaps** (1–3 bullets, skip if none): Use format: "Has [qualifier] [skill]; role requires [qualifier] [skill]". Be precise. | |
| **Flags** (only if flags list is non-empty): Briefly explain recruiter concern for each flag. | |
| **Recommendation**: {tier} | |
| Be specific. No generic statements. Reference the actual data provided.""" | |
| def _fallback_explanation(candidate_detail: dict, tier: str) -> str: | |
| name = candidate_detail.get("candidate_name", "Candidate") | |
| position = candidate_detail.get("current_position", "") | |
| years = candidate_detail.get("years_of_experience", 0) | |
| score = candidate_detail.get("final_score", 0) | |
| breakdown = candidate_detail.get("score_breakdown", {}) | |
| if isinstance(breakdown, str): | |
| try: | |
| breakdown = json.loads(breakdown) | |
| except Exception: | |
| breakdown = {} | |
| matched = breakdown.get("matched_skills", []) | |
| if isinstance(matched, str): | |
| try: | |
| matched = json.loads(matched) | |
| except Exception: | |
| matched = [] | |
| matched_names = [s.get("skill", s) if isinstance(s, dict) else s for s in matched[:5]] | |
| missing = breakdown.get("missing_skills", []) | |
| if isinstance(missing, str): | |
| try: | |
| missing = json.loads(missing) | |
| except Exception: | |
| missing = [] | |
| strengths = ", ".join(matched_names[:3]) if matched_names else "relevant technical skills" | |
| gaps_text = f"- Missing or below-level: {', '.join(missing[:3])}" if missing else "- No critical gaps identified" | |
| return f"""**Overall Fit** | |
| {name} is a {tier.lower()} for this role with a {score:.0%} match score. Currently working as {position} with {years if years != -1 else 'unknown'} years of experience. | |
| **Strengths** | |
| - Strong in: {strengths} | |
| - {years if years != -1 else 'Experience'} years of relevant background | |
| - Current role aligns with core requirements | |
| **Gaps** | |
| {gaps_text} | |
| **Recommendation**: {tier} | |
| *Note: Detailed AI explanation unavailable — Ollama not running. Start Ollama and pull the model to enable.*""" | |
| def generate_explanation(jd_data: dict, candidate_detail: dict) -> str: | |
| """ | |
| Generate a human-readable explanation via Ollama cloud. | |
| Falls back to template if Ollama is unavailable. | |
| """ | |
| from app.config import settings | |
| tier = candidate_detail.get("tier", "Partial Match") | |
| try: | |
| prompt = _build_prompt(jd_data, candidate_detail) | |
| client = _get_client() | |
| response = client.chat( | |
| model=settings.ollama_model, | |
| messages=[{"role": "user", "content": prompt}], | |
| options={"temperature": 0.2, "num_predict": 1500}, | |
| ) | |
| text = response.message.content | |
| return text.strip() | |
| except Exception as e: | |
| logger.warning(f"Ollama call failed: {e}, using fallback") | |
| return _fallback_explanation(candidate_detail, tier) |