diff --git a/yanfr-lab-open-hr/.env.example b/yanfr-lab-open-hr/.env.example new file mode 100644 index 0000000..4d89586 --- /dev/null +++ b/yanfr-lab-open-hr/.env.example @@ -0,0 +1,4 @@ +HRFLOW_API_KEY=your_api_key_here +HRFLOW_USER_EMAIL=your_email_here +HRFLOW_SOURCE_KEY=your_source_key_here +HRFLOW_BOARD_KEY=your_board_key_here diff --git a/yanfr-lab-open-hr/README.md b/yanfr-lab-open-hr/README.md new file mode 100644 index 0000000..a2b8b06 --- /dev/null +++ b/yanfr-lab-open-hr/README.md @@ -0,0 +1,59 @@ +# Open HR avec HrFlow.ai + +> Intelligent candidate sourcing with auto-generated questionnaires and HrFlow.ai-powered scoring. + +## What it does + +Open HR is an AI sourcing agent built for recruiters. Given a job title and requirements, it: + +1. **Auto-generates a weighted questionnaire** from the job description (skills, experience, education, stability) +2. **Indexes the job** via HrFlow.ai Job Indexing API +3. **Scores all candidates** in the source using HrFlow.ai Scoring API with a custom algorithm key +4. **Combines AI scores with questionnaire weights** to produce precise decimal rankings +5. **Allows refinement** via natural language feedback loop + +## HrFlow.ai APIs used + +- `POST /v1/job/indexing` — Index job description to enable scoring +- `GET /v1/profiles/scoring` — Score all profiles against the job using algorithm key +- `GET /v1/profiles/searching` — Fallback profile search +- `POST /v1/profile/indexing` — Index candidate profiles + +## How to run + +### Prerequisites + +- Python 3.11+ + +### Setup + +```bash +# Install dependencies +pip install -r requirements.txt + +# Copy environment variables +cp .env.example .env +# Fill in your HrFlow API keys in .env + +# Start the app +uvicorn web_app_fr_v2:app --host 0.0.0.0 --port 8002 +``` + +### Environment variables + +| Variable | Required | Description | +|----------|----------|-------------| +| `HRFLOW_API_KEY` | Yes | HrFlow.ai API secret key | +| `HRFLOW_USER_EMAIL` | Yes | HrFlow.ai account email | +| `HRFLOW_SOURCE_KEY` | Yes | HrFlow.ai source key | +| `HRFLOW_BOARD_KEY` | Yes | HrFlow.ai board key | + +## Screenshots + +![Preview](./assets/preview.png) + +## Team + +- **Yan** — Full-stack development & HrFlow.ai integration +- **Leo** — HrFlow.ai API integration & scoring algorithm +- **Aike** — Development & HrFlow.ai integration diff --git a/yanfr-lab-open-hr/app.json b/yanfr-lab-open-hr/app.json new file mode 100644 index 0000000..171322d --- /dev/null +++ b/yanfr-lab-open-hr/app.json @@ -0,0 +1,16 @@ +{ + "$schema": "../../schemas/app.schema.json", + "name": "Open HR avec HrFlow.ai", + "description": "AI agent that auto-generates a weighted questionnaire from any job description, scores and ranks candidates using HrFlow.ai Scoring API, and refines results through a natural language feedback loop.", + "credentials": { + "source_keys": ["c1039b004a836e42427b9c89309ad99cf9b6a73c"], + "board_keys": ["c99d70c3a062c2f99ca78fce23b89fb4f53119ef"], + "algorithm_key": "b1ebac4c62fa96e06206f4433b95ae69674891ff" + }, + "settings": { + "team_name": "yanfr-lab", + "theme_color": "#83D9DC", + "custom_filters": [], + "filters": [] + } +} diff --git a/yanfr-lab-open-hr/assets/preview.png b/yanfr-lab-open-hr/assets/preview.png new file mode 100755 index 0000000..dad7221 Binary files /dev/null and b/yanfr-lab-open-hr/assets/preview.png differ diff --git a/yanfr-lab-open-hr/database.py b/yanfr-lab-open-hr/database.py new file mode 100644 index 0000000..9dc9688 --- /dev/null +++ b/yanfr-lab-open-hr/database.py @@ -0,0 +1,74 @@ +import sqlite3 +import json +from datetime import datetime + +DB_NAME = "sourcing.db" + +def init_db(): + conn = sqlite3.connect(DB_NAME) + cursor = conn.cursor() + cursor.execute(''' + CREATE TABLE IF NOT EXISTS searches ( + id INTEGER PRIMARY KEY AUTOINCREMENT, + query TEXT, + created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP + ) + ''') + cursor.execute(''' + CREATE TABLE IF NOT EXISTS candidates ( + id INTEGER PRIMARY KEY AUTOINCREMENT, + search_id INTEGER, + name TEXT, + headline TEXT, + location TEXT, + profile_url TEXT, + source_platform TEXT, + hrflow_profile_key TEXT, + score REAL, + rank INTEGER, + raw_json TEXT, + created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP, + FOREIGN KEY(search_id) REFERENCES searches(id) + ) + ''') + conn.commit() + conn.close() + +def save_search(query: str) -> int: + conn = sqlite3.connect(DB_NAME) + cursor = conn.cursor() + cursor.execute("INSERT INTO searches (query) VALUES (?)", (query,)) + search_id = cursor.lastrowid + conn.commit() + conn.close() + return search_id + +def save_candidate(search_id: int, candidate: dict): + conn = sqlite3.connect(DB_NAME) + cursor = conn.cursor() + cursor.execute(''' + INSERT INTO candidates (search_id, name, headline, location, profile_url, + source_platform, hrflow_profile_key, score, rank, raw_json) + VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?) + ''', ( + search_id, candidate['name'], candidate.get('headline'), candidate.get('location'), + candidate['profile_url'], candidate.get('source_platform', 'github'), + candidate.get('hrflow_profile_key'), candidate.get('score', 0), + candidate.get('rank'), json.dumps(candidate) + )) + conn.commit() + conn.close() + +def get_recent_searches(limit=20): + conn = sqlite3.connect(DB_NAME) + conn.row_factory = sqlite3.Row + cursor = conn.cursor() + cursor.execute(''' + SELECT s.*, (SELECT COUNT(*) FROM candidates WHERE search_id = s.id) as candidate_count + FROM searches s ORDER BY created_at DESC LIMIT ? + ''', (limit,)) + rows = [dict(row) for row in cursor.fetchall()] + conn.close() + return rows + +init_db() \ No newline at end of file diff --git a/yanfr-lab-open-hr/hrflow/__init__.py b/yanfr-lab-open-hr/hrflow/__init__.py new file mode 100644 index 0000000..e69de29 diff --git a/yanfr-lab-open-hr/hrflow/client.py b/yanfr-lab-open-hr/hrflow/client.py new file mode 100644 index 0000000..2eda25a --- /dev/null +++ b/yanfr-lab-open-hr/hrflow/client.py @@ -0,0 +1,537 @@ +""" +HrFlow AI Client — 12 APIs，使用已验证的正确端点。 + +Base URL: https://api.hrflow.ai/v1 + +APIs: + 1. Parsing API — POST /v1/text/parsing + POST /v1/job/indexing + 2. Tagging API — POST /v1/text/tagging + 3. Embedding API — POST /v1/profile/embedding + 4. Searching API — GET /v1/profiles/searching + 5. Matching API — GET /v1/profiles/scoring (无阈值宽召回) + 6. Scoring API — GET /v1/profiles/scoring (有阈值精排) + 7. Grading API — GET /v1/profile/grading + 8. Reasoning API — GET /v1/profile/revealing + 9. Signals API — POST /v1/profile/events + 10. Upskilling API — GET /v1/jobs/searching (skill gap) + 11. Data Studio — POST /v1/profile/indexing (batch sync) + 12. UI Studio — GET /v1/profile/indexing (card data) +""" + +from __future__ import annotations + +import io +import json +import logging +from typing import Any, Optional + +import httpx +from tenacity import ( + retry, retry_if_exception_type, + stop_after_attempt, wait_exponential, +) + +from config import config + +logger = logging.getLogger(__name__) +TIMEOUT = httpx.Timeout(30.0) +BASE = "https://api.hrflow.ai/v1" + + +def _auth_headers() -> dict: + return { + "X-API-KEY": config.HRFLOW_API_KEY, + "X-USER-EMAIL": config.HRFLOW_USER_EMAIL, + } + + +def _check(resp: httpx.Response, label: str) -> dict: + if resp.status_code not in (200, 201): + logger.error("[HrFlow/%s] HTTP %s: %s", label, resp.status_code, resp.text[:300]) + resp.raise_for_status() + body = resp.json() + if body.get("code") not in (200, 201, None): + raise RuntimeError(f"[HrFlow/{label}] error {body.get('code')}: {body.get('message')}") + return body + + +def _get(path: str, params: dict) -> dict: + with httpx.Client(timeout=TIMEOUT) as c: + r = c.get(f"{BASE}{path}", headers={**_auth_headers(), "Content-Type": "application/json"}, params=params) + return _check(r, path) + + +def _post_json(path: str, payload: dict) -> dict: + with httpx.Client(timeout=TIMEOUT) as c: + r = c.post(f"{BASE}{path}", headers={**_auth_headers(), "Content-Type": "application/json"}, json=payload) + return _check(r, path) + + +def _post_multipart(path: str, data: dict, files: dict) -> dict: + with httpx.Client(timeout=TIMEOUT) as c: + r = c.post(f"{BASE}{path}", headers=_auth_headers(), data=data, files=files) + return _check(r, path) + + +class HrFlowClient: + + # ───────────────────────────────────────────────────────────────────────── + # 1. PARSING API + # ───────────────────────────────────────────────────────────────────────── + + @retry(stop=stop_after_attempt(3), wait=wait_exponential(min=2, max=10), + retry=retry_if_exception_type((httpx.HTTPError, RuntimeError))) + def parse_job_from_text(self, board_key: str, job_text: str, job_title: str) -> dict: + """ + Parsing API (Job) — POST /v1/job/indexing + 直接将 JD 写入 Board 获取 job_key（text_parsing 为可选增强）。 + """ + import uuid + job_key = str(uuid.uuid4()) + + # 尝试用 text/parsing 提取技能（若账号未开通则跳过） + skills = [] + try: + parse_resp = _post_json("/text/parsing", {"text": f"{job_title}\n\n{job_text}"}) + skills = parse_resp.get("data", {}).get("skills", []) + except Exception as e: + logger.info("[Parsing] text/parsing skipped (not available on this plan): %s", e) + + job_payload = { + "board_key": board_key, + "job": { + "key": job_key, + "name": job_title, + "description": job_text, + "skills": skills, + "tags": [], + "location": {"text": ""}, + "sections": [{"name": "description", "title": "Description", "description": job_text}], + }, + } + index_resp = _post_json("/job/indexing", job_payload) + # 返回 job_key 合并到响应中 + index_resp["_job_key"] = job_key + return index_resp + + @retry(stop=stop_after_attempt(3), wait=wait_exponential(min=2, max=10), + retry=retry_if_exception_type((httpx.HTTPError, RuntimeError))) + def parse_profile_from_url( + self, resume_url: str, source_key: str, texts: Optional[list[str]] = None + ) -> dict: + """ + Parsing API (Profile) — POST /v1/profile/parsing/file + 下载简历文件后以 multipart 上传。 + 若 URL 不可访问，退回到直接索引一个最小 profile。 + texts: GitHub README 等补充文本，拼接后附加到简历内容。 + """ + # 尝试下载简历 + file_bytes: Optional[bytes] = None + filename = "resume.pdf" + try: + with httpx.Client(timeout=httpx.Timeout(15.0)) as c: + dl = c.get(resume_url, follow_redirects=True) + if dl.status_code == 200: + file_bytes = dl.content + # 从 URL 取文件名 + filename = resume_url.split("/")[-1] or "resume.pdf" + except Exception as e: + logger.warning("[Parsing] Resume download failed (%s): %s", resume_url, e) + + if file_bytes: + # 若有补充文本，拼成额外 txt 一起上传 + extra = ("\n\n".join(texts) if texts else "").encode("utf-8") + if extra: + with httpx.Client(timeout=TIMEOUT) as c: + r = c.post( + f"{BASE}/profile/parsing/file", + headers=_auth_headers(), + data={"source_key": source_key, "sync_parsing": "1"}, + files={ + "file": (filename, file_bytes, "application/pdf"), + "file2": ("github.txt", extra, "text/plain"), + }, + ) + return _check(r, "/profile/parsing/file") + else: + return _post_multipart( + "/profile/parsing/file", + data={"source_key": source_key, "sync_parsing": "1"}, + files={"file": (filename, file_bytes, "application/pdf")}, + ) + else: + # 无法下载 → 直接用文本索引一个 profile + combined_text = f"{resume_url}\n\n" + ("\n\n".join(texts) if texts else "") + return self._index_profile_from_text(source_key, combined_text) + + def _index_profile_from_text(self, source_key: str, text: str) -> dict: + """兜底：当简历 URL 不可访问时，直接用文本解析后索引 profile。""" + import uuid + profile_key = str(uuid.uuid4()) + # 先用 text/parsing 提取结构化信息 + try: + parsed = _post_json("/text/parsing", {"text": text[:3000]}) + skills = parsed.get("data", {}).get("skills", []) + except Exception: + skills = [] + + payload = { + "source_key": source_key, + "profile": { + "key": profile_key, + "skills": skills, + "experiences": [], + "educations": [], + "languages": [], + "certifications": [], + "info": {"full_name": "Unknown", "summary": text[:500]}, + "sections": [{"name": "raw", "title": "Raw", "description": text[:2000]}], + }, + } + resp = _post_json("/profile/indexing", payload) + resp["_profile_key"] = profile_key + return resp + + # ───────────────────────────────────────────────────────────────────────── + # 2. TAGGING API — POST /v1/text/tagging + # ───────────────────────────────────────────────────────────────────────── + + def tag_profile(self, profile_key: str, source_key: str) -> dict: + """ + Tagging API — 取 profile 文本后调用 POST /v1/text/tagging + 用 hrflow-skills tagger 打 HR 分类标签，统一技能词汇表。 + """ + # 先取 profile 内容 + try: + p = _get("/profile/indexing", {"source_key": source_key, "profile_key": profile_key}) + sections = p.get("data", {}).get("profile", {}).get("sections", []) + text = " ".join(s.get("description", "") for s in sections)[:2000] or profile_key + except Exception: + text = profile_key + + try: + return _post_json("/text/tagging", { + "text": text, + "algorithm_key": "tagger-hrflow-skills", + "top_n": 10, + "output_lang": "en", + }) + except Exception as e: + logger.info("[Tagging] text/tagging skipped (not available): %s", e) + return {"code": 200, "data": {"tags": []}} + + def tag_job(self, board_key: str, job_key: str) -> dict: + """Tagging API — 对 Job 文本打 HR 技能标签。""" + try: + j = _get("/job/indexing", {"board_key": board_key, "job_key": job_key}) + desc = j.get("data", {}).get("job", {}).get("description", "") or job_key + except Exception: + desc = job_key + + try: + return _post_json("/text/tagging", { + "text": desc[:2000], + "algorithm_key": "tagger-hrflow-skills", + "top_n": 10, + "output_lang": "en", + }) + except Exception as e: + logger.info("[Tagging] text/tagging skipped (not available): %s", e) + return {"code": 200, "data": {"tags": []}} + + # ───────────────────────────────────────────────────────────────────────── + # 3. EMBEDDING API — POST /v1/profile/embedding + # ───────────────────────────────────────────────────────────────────────── + + def embed_profile(self, profile_key: str, source_key: str) -> dict: + """ + Embedding API — POST /v1/profile/embedding + 生成语义向量，为 Searching / Scoring 提供语义相似度基础。 + """ + return _post_json("/profile/embedding", { + "source_key": source_key, + "profile_key": profile_key, + }) + + def embed_job(self, board_key: str, job_key: str) -> dict: + """Embedding API — POST /v1/job/embedding 生成 Job 语义向量。""" + return _post_json("/job/embedding", { + "board_key": board_key, + "job_key": job_key, + }) + + # ───────────────────────────────────────────────────────────────────────── + # 4. SEARCHING API — GET /v1/profiles/searching + # ───────────────────────────────────────────────────────────────────────── + + def search_profiles( + self, + source_keys: list[str], + query: str, + filters: Optional[dict] = None, + page: int = 1, + limit: int = 50, + ) -> dict: + """ + Searching API — GET /v1/profiles/searching + 关键词+语义混合检索，作为 Scoring 前的轻量预过滤，降低成本。 + """ + params: dict[str, Any] = { + "source_keys": json.dumps(source_keys), + "text_keywords": query, + "page": page, + "limit": limit, + "sort_by": "searching", + "order_by": "desc", + } + if filters: + params.update(filters) + return _get("/profiles/searching", params) + + # ───────────────────────────────────────────────────────────────────────── + # 5. MATCHING API — GET /v1/profiles/scoring (无阈值宽召回) + # ───────────────────────────────────────────────────────────────────────── + + def match_profiles_for_job( + self, + job_key: str, + board_key: str, + source_keys: list[str], + limit: int = 100, + ) -> dict: + """ + Matching API — GET /v1/profiles/scoring (无 score_threshold) + 粗粒度宽匹配，扩大召回范围后再由 Scoring API 精排。 + """ + return _get("/profiles/scoring", { + "job_key": job_key, + "board_key": board_key, + "source_keys": json.dumps(source_keys), + "limit": limit, + "sort_by": "scoring", + "order_by": "desc", + }) + + # ───────────────────────────────────────────────────────────────────────── + # 6. SCORING API — GET /v1/profiles/scoring (核心) + # ───────────────────────────────────────────────────────────────────────── + + def score_profiles_for_job( + self, + job_key: str, + board_key: str, + source_keys: list[str], + score_threshold: float = 0.6, + limit: int = 20, + page: int = 1, + use_memory: bool = False, + ) -> dict: + """ + Scoring API — GET /v1/profiles/scoring + 返回 predictions[1] 概率分（0–1），按 score desc 排序。 + use_memory=True 激活在线微调（需先有 Signals）。 + """ + params: dict[str, Any] = { + "job_key": job_key, + "board_key": board_key, + "source_keys": json.dumps(source_keys), + "limit": limit, + "page": page, + "sort_by": "scoring", + "order_by": "desc", + } + if use_memory: + params["use_memory_context"] = 1 + return _get("/profiles/scoring", params) + + def index_job_sync(self, board_key: str, job_title: str) -> str: + """ + Indexing API (Job) — POST /v1/job/indexing + Crée un job dans le Board et retourne son job_key. + Approche Leo : job minimal pour déclencher le scoring. + """ + import time as _time + payload = { + "board_key": board_key, + "name": job_title[:50], + "reference": f"ref-{int(_time.time())}", + "summary": job_title, + "location": {"text": "Remote", "lat": None, "lng": None}, + "sections": [{"name": "description", "title": "Job Description", "description": job_title}], + } + resp = _post_json("/job/indexing", payload) + return resp["data"]["key"] + + def score_profiles_with_algo( + self, + job_key: str, + board_key: str, + source_keys: list[str], + limit: int = 100, + ) -> dict: + """ + Scoring API — GET /v1/profiles/scoring avec algorithm_key (approche Leo). + Retourne profiles + predictions[][1] = score IA (0–1). + """ + params: dict[str, Any] = { + "job_key": job_key, + "board_key": board_key, + "source_keys": json.dumps(source_keys), + "algorithm_key": "b1ebac4c62fa96e06206f4433b95ae69674891ff", + "limit": limit, + "sort_by": "scoring", + "order_by": "desc", + } + with httpx.Client(timeout=TIMEOUT) as c: + r = c.get(f"{BASE}/profiles/scoring", + headers={**_auth_headers(), "Content-Type": "application/json"}, + params=params) + if r.status_code != 200: + logger.error("[HrFlow/scoring-algo] HTTP %s: %s", r.status_code, r.text[:300]) + return {} + return r.json() + + # ───────────────────────────────────────────────────────────────────────── + # 7. GRADING API — GET /v1/profile/grading + # ───────────────────────────────────────────────────────────────────────── + + def grade_profile( + self, profile_key: str, source_key: str, job_key: str, board_key: str + ) -> dict: + """ + Grading API — GET /v1/profile/grading + 对单个 profile-job 对进行二次重排，输出等级（A/B/C）。 + 在 Scoring Top-N 之后调用，进一步精细筛选。 + """ + return _get("/profile/grading", { + "source_key": source_key, + "profile_key": profile_key, + "board_key": board_key, + "job_key": job_key, + }) + + # ───────────────────────────────────────────────────────────────────────── + # 8. REASONING API — GET /v1/profile/revealing + # ───────────────────────────────────────────────────────────────────────── + + def get_reasoning( + self, profile_key: str, source_key: str, job_key: str, board_key: str + ) -> dict: + """ + Reasoning API — GET /v1/profile/revealing + 返回评分依据的自然语言解释，满足 EU AI Act 可解释性要求。 + """ + return _get("/profile/revealing", { + "source_key": source_key, + "profile_key": profile_key, + "board_key": board_key, + "job_key": job_key, + }) + + # ───────────────────────────────────────────────────────────────────────── + # 9. SIGNALS API — POST /v1/profile/events + # ───────────────────────────────────────────────────────────────────────── + + def send_signal( + self, + profile_key: str, + source_key: str, + job_key: str, + board_key: str, + event_type: str, + rating: Optional[float] = None, + ) -> dict: + """ + Signals API — POST /v1/profile/events + 将招聘官动作（面试/拒绝/offer）回写给 HrFlow， + 驱动在线微调，使后续评分越来越准确。 + """ + payload: dict[str, Any] = { + "source_key": source_key, + "profile_key": profile_key, + "board_key": board_key, + "job_key": job_key, + "type": event_type, + } + if rating is not None: + payload["rating"] = rating + return _post_json("/profile/events", payload) + + # ───────────────────────────────────────────────────────────────────────── + # 10. UPSKILLING API — GET /v1/jobs/searching (skill gap) + # ───────────────────────────────────────────────────────────────────────── + + def get_upskilling( + self, profile_key: str, source_key: str, job_key: str, board_key: str + ) -> dict: + """ + Upskilling API — GET /v1/profile/indexing (比对技能缺口) + 获取 profile 和 job 的技能列表，计算 gap（job 有而 profile 缺少的技能）。 + 结果写入候选人输出卡片的"技能缺口"字段。 + """ + # 取 profile 技能 + try: + p_resp = _get("/profile/indexing", { + "source_key": source_key, + "profile_key": profile_key, + }) + p_skills = {s.get("name", "").lower() + for s in p_resp.get("data", {}).get("profile", {}).get("skills", [])} + except Exception: + p_skills = set() + + # 取 job 技能 + try: + j_resp = _get("/job/indexing", { + "board_key": board_key, + "job_key": job_key, + }) + j_skills = [s for s in j_resp.get("data", {}).get("job", {}).get("skills", [])] + except Exception: + j_skills = [] + + # 计算缺口 + gaps = [s for s in j_skills if s.get("name", "").lower() not in p_skills] + return {"code": 200, "data": {"upskilling": gaps}} + + # ───────────────────────────────────────────────────────────────────────── + # 11. DATA STUDIO — POST /v1/profile/indexing (批量同步) + # ───────────────────────────────────────────────────────────────────────── + + def sync_profiles_to_source(self, source_key: str, profiles: list[dict]) -> dict: + """ + Data Studio — 批量将候选人 profile 写入 HrFlow Source。 + 生产环境中由 Data Studio 的 200+ 连接器（Greenhouse/Lever/Workday）替代。 + """ + results = [] + for p in profiles: + try: + r = _post_json("/profile/indexing", { + "source_key": source_key, + "profile": p, + }) + results.append({"key": p.get("key"), "status": "ok"}) + except Exception as e: + results.append({"key": p.get("key"), "status": "error", "message": str(e)}) + return {"code": 200, "data": {"synced": results}} + + def list_sources(self) -> dict: + """Data Studio — GET /v1/sources 列出所有数据源。""" + return _get("/sources", {}) + + # ───────────────────────────────────────────────────────────────────────── + # 12. UI STUDIO — GET /v1/profile/indexing (recruiter card data) + # ───────────────────────────────────────────────────────────────────────── + + def get_ui_profile_card( + self, profile_key: str, source_key: str, job_key: str, board_key: str + ) -> dict: + """ + UI Studio — GET /v1/profile/indexing + 返回结构化 profile 数据，供前端招聘官卡片组件渲染。 + 包含匿名化字段、技能高亮和匹配指示器。 + """ + return _get("/profile/indexing", { + "source_key": source_key, + "profile_key": profile_key, + }) diff --git a/yanfr-lab-open-hr/hrflow_client.py b/yanfr-lab-open-hr/hrflow_client.py new file mode 100644 index 0000000..40002f8 --- /dev/null +++ b/yanfr-lab-open-hr/hrflow_client.py @@ -0,0 +1,93 @@ +import os +import httpx +import hashlib +import time +from dotenv import load_dotenv +import asyncio + +load_dotenv() + +API_KEY = os.getenv("HRFLOW_API_KEY") +USER_EMAIL = os.getenv("HRFLOW_USER_EMAIL") +SOURCE_KEY = os.getenv("HRFLOW_SOURCE_KEY") +BOARD_KEY = os.getenv("HRFLOW_BOARD_KEY") + +HEADERS = {"X-API-KEY": API_KEY, "X-USER-EMAIL": USER_EMAIL, "Content-Type": "application/json"} +BASE_URL = "https://api.hrflow.ai/v1" + + +async def index_job(query: str) -> str: + async with httpx.AsyncClient() as client: + import time + payload = { + "board_key": BOARD_KEY, + "name": query[:50], + "reference": f"ref-{int(time.time())}", + "summary": query, + "location": { + "text": "Remote", # 必须包含此字段 + "lat": None, + "lng": None + }, + "sections": [ + { + "name": "description", + "title": "Job Description", + "description": query + } + ] + } + + print(f"正在请求 HrFlow API... Board Key: {BOARD_KEY}") + + resp = await client.post( + f"{BASE_URL}/job/indexing", + json=payload, + headers=HEADERS + ) + + if resp.status_code not in [200, 201]: + print("--- HrFlow API 返回了错误 ---") + print(f"状态码: {resp.status_code}") + print(f"详细错误内容: {resp.text}") + print("----------------------------") + raise Exception(f"HrFlow 报错: {resp.text}") + + return resp.json()["data"]["key"] + +async def index_profile(candidate: dict) -> str: + ref = hashlib.md5(candidate['profile_url'].encode()).hexdigest()[:16] + payload = {"source_key": SOURCE_KEY, "reference": ref, "info": {"full_name": candidate['name'], "location": {"text": candidate['location']}, "urls": [{"type": "from_resume", "url": candidate['profile_url']}], "summary": candidate['headline'], "picture": candidate['avatar_url']}, "text": candidate['raw_text']} + async with httpx.AsyncClient() as client: + resp = await client.post(f"{BASE_URL}/profile/indexing", json=payload, headers=HEADERS) + return resp.json()["data"]["key"] if resp.status_code != 409 else resp.json()["data"]["key"] + + +async def score_profiles(job_key: str, limit: int = 10): + params = { + "board_key": BOARD_KEY, + "job_key": job_key, + "source_keys": f'["{SOURCE_KEY}"]', # 必须是这种 ["key"] 的格式 + "algorithm_key": 'b1ebac4c62fa96e06206f4433b95ae69674891ff', + "limit": limit, + "sort_by": "scoring", + "order_by": "desc" + } + + async with httpx.AsyncClient(timeout=30.0) as client: + resp = await client.get(f"{BASE_URL}/profiles/scoring", params=params, headers=HEADERS) + + if resp.status_code == 200: + # --- 关键：返回整个 JSON 字典 --- + return resp.json() + else: + print(f"❌ API报错: {resp.status_code}") + return {} + +async def explain_profile(job_key: str, profile_key: str) -> str: + params = {"board_key": BOARD_KEY, "job_key": job_key, "source_key": SOURCE_KEY, "profile_key": profile_key} + async with httpx.AsyncClient() as client: + try: + resp = await client.get(f"{BASE_URL}/profile/explaining", params=params, headers=HEADERS, timeout=2.0) + return resp.json()["data"]["explanation"] + except: return "" \ No newline at end of file diff --git a/yanfr-lab-open-hr/main.py b/yanfr-lab-open-hr/main.py new file mode 100644 index 0000000..d0cf76e --- /dev/null +++ b/yanfr-lab-open-hr/main.py @@ -0,0 +1,79 @@ +import asyncio +from fastapi import FastAPI, Request +from fastapi.responses import HTMLResponse +from fastapi.staticfiles import StaticFiles +import database as db +import sourcing +import hrflow_client as hrflow +import uvicorn +import os +app = FastAPI() +if not os.path.exists("static"): os.makedirs("static") +app.mount("/static", StaticFiles(directory="static"), name="static") + +@app.get("/", response_class=HTMLResponse) +async def read_index(): + with open("static/index.html", encoding="utf-8") as f: return f.read() + +@app.post("/search") +@app.post("/search") +async def perform_search(request: Request): + try: + data = await request.json() + query = data.get("query") + if not query: + return {"status": "error", "message": "请输入搜索关键词"} + + print(f"🔍 正在为岗位创建索引: {query}") + # 1. 创建 Job (这一步会让 Board Key 计数) + job_key = await hrflow.index_job(query) + + # 2. 给 AI 一点反应时间 (因为没有了爬虫过程，建议保留几秒等待) + print("⏳ 正在请求 HrFlow AI 计算匹配得分...") + await asyncio.sleep(5) + + # 3. 直接获取评分结果 (搜索 Source 中已有的所有人) + full_response = await hrflow.score_profiles(job_key) + + # 解析数据 + api_data = full_response.get('data', {}) + profiles = api_data.get('profiles', []) + predictions = api_data.get('predictions', []) + + results = [] + for i, profile in enumerate(profiles): + info = profile.get('info', {}) + + # 提取姓名 + display_name = info.get('full_name') or profile.get('reference') or "未知候选人" + + # 提取分数 + score_percent = "0%" + if i < len(predictions) and len(predictions[i]) > 1: + score_percent = f"{round(predictions[i][1] * 100, 2)}%" + + # 提取链接 + link = "#" + for u in info.get('urls', []): + if isinstance(u, dict) and u.get('url'): + link = u['url'] + break + + results.append({ + "name": display_name, + "score": score_percent, + "url": link, + "summary": info.get('summary') or "暂无个人简介" + }) + + print(f"✅ 成功找到 {len(results)} 位匹配的人才") + return {"status": "success", "results": results} + + except Exception as e: + print(f"❌ 搜索出错: {str(e)}") + return {"status": "error", "message": str(e)} +@app.get("/history") +async def get_history(): return db.get_recent_searches() + +if __name__ == "__main__": + uvicorn.run(app, host="0.0.0.0", port=8000) \ No newline at end of file diff --git a/yanfr-lab-open-hr/requirements.txt b/yanfr-lab-open-hr/requirements.txt new file mode 100644 index 0000000..368eae4 --- /dev/null +++ b/yanfr-lab-open-hr/requirements.txt @@ -0,0 +1,4 @@ +fastapi==0.111.0 +uvicorn==0.29.0 +httpx==0.27.0 +python-dotenv==1.0.1 \ No newline at end of file diff --git a/yanfr-lab-open-hr/sourcing.py b/yanfr-lab-open-hr/sourcing.py new file mode 100644 index 0000000..8da3dfb --- /dev/null +++ b/yanfr-lab-open-hr/sourcing.py @@ -0,0 +1,46 @@ +import httpx +import asyncio +import re + + +async def search_candidates(query: str, n: int = 30) -> list[dict]: + query_lower = query.lower() + role = "developer" + skills = [] + + if "python" in query_lower: skills.append("python") + if "react" in query_lower or "frontend" in query_lower: + skills.append("react") + role = "frontend-developer" + + search_queries = [f"{role} {' '.join(skills)}", f"{role} engineer"] + all_candidates = {} + + async with httpx.AsyncClient(timeout=10.0) as client: + for sq in search_queries: + try: + url = f"https://api.github.com/search/users?q={sq}+in:bio+type:user&per_page=15" + headers = {"Accept": "application/vnd.github+json", "User-Agent": "SourcingAgent/1.0"} + resp = await client.get(url, headers=headers) + if resp.status_code != 200: continue + + for item in resp.json().get("items", []): + login = item["login"] + if login in all_candidates: continue + await asyncio.sleep(0.1) + u_resp = await client.get(f"https://api.github.com/users/{login}", headers=headers) + if u_resp.status_code == 200: + u = u_resp.json() + all_candidates[login] = { + "name": u.get("name") or u.get("login"), + "headline": (u.get("bio") or "")[:200], + "location": u.get("location") or "Remote", + "profile_url": u["html_url"], + "source_platform": "github", + "avatar_url": u.get("avatar_url"), + "company": u.get("company") or "", + "raw_text": f"{u.get('name')} {u.get('bio')} {u.get('location')}" + } + except: pass + if len(all_candidates) >= n: break + return list(all_candidates.values())[:n] \ No newline at end of file diff --git a/yanfr-lab-open-hr/static/index_openhr.html b/yanfr-lab-open-hr/static/index_openhr.html new file mode 100644 index 0000000..312272b --- /dev/null +++ b/yanfr-lab-open-hr/static/index_openhr.html @@ -0,0 +1,691 @@ + + + + + +Open HR + + + + + +
+
+ +
+

Chargement

+
+ +
+
+ +
+

+ HrFlow + Open HR +

+

Sourcing intelligent · Questionnaire adaptatif · Pondération par critère · Boucle d'affinement IA

+
+ +
+
1
Poste
+
2
Questionnaire
+
3
Affinement
+
4
Résultats
+
+ +
+
+ + +
+
+

📋 Décrivez le poste à pourvoir

+
+ + +
+ +
+
+ + +
+
+

🎯 Précisez vos critères de recrutement

+

+ Répondez aux questions et ajustez le curseur de pondération à droite de chaque question. + Plus le curseur est à droite, plus ce critère pèse dans le scoring final. +

+
+ Poids : + 1 = indicatif + · + 3 = standard + · + 5 = décisif +
+
+
Questions spécifiques au poste
+
+
+ + +
+
+
+ + +
+
+
+
+ 🤖 Agent Open HR : Plus de 10 candidats correspondent à vos critères actuels. + Pour affiner la sélection, décrivez un critère supplémentaire ou une préférence spécifique. +

+ Ex : « doit maîtriser Kubernetes », « minimum 7 ans d'expérience », « master requis », « profil stable »… +
+
+ + +
+
+
+ + +
+
+ + +
+
+
+
+
+ + +
+
+
+ + + + + + diff --git a/yanfr-lab-open-hr/web_app_fr_v2.py b/yanfr-lab-open-hr/web_app_fr_v2.py new file mode 100644 index 0000000..d2dd40c --- /dev/null +++ b/yanfr-lab-open-hr/web_app_fr_v2.py @@ -0,0 +1,962 @@ +""" +AI Agent avec OpenClaw — Version 2 +Nouveautés v2 : + - Question de stabilité des employés dans le questionnaire général + - Poids dynamiques par question (slider côté frontend → envoyés au backend) + - Évaluation de la stabilité des candidats (durée moyenne de tenure) +Port 8002. +""" + +import json +import os +import re +import sys +import time +import uuid +from collections import Counter +from datetime import datetime +from typing import Optional + +from fastapi import FastAPI, HTTPException +from fastapi.middleware.cors import CORSMiddleware +from fastapi.responses import FileResponse +from fastapi.staticfiles import StaticFiles +from pydantic import BaseModel + +sys.path.insert(0, os.path.dirname(__file__)) + +from config import config +from hrflow.client import HrFlowClient + +app = FastAPI(title="AI Agent avec OpenClaw v2") +app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"]) +app.mount("/static", StaticFiles(directory="static"), name="static") + +sessions: dict = {} +hrflow = HrFlowClient() + + +# ─── Request Models ─────────────────────────────────────────────────────────── + +class StartRequest(BaseModel): + job_title: str + job_description: str + +class SubmitAnswersRequest(BaseModel): + session_id: str + answers: dict + weights: dict = {} # { question_id: slider_value (1–5) } + +class RefineRequest(BaseModel): + session_id: str + refinement: str + + +# ─── Catégorie de poste ──────────────────────────────────────────────────────── + +CATEGORY_SIGNALS = { + "frontend": ["frontend","front-end","前端","react","vue","angular","svelte","css","html", + "webpack","vite","nextjs","nuxt","ui/ux","界面","interface","intégration"], + "backend": ["backend","back-end","后端","服务端","api","server","django","flask","fastapi", + "spring","rails","express","nestjs","grpc","microservice","微服务","serveur"], + "fullstack": ["fullstack","full-stack","全栈","full stack","complet","polyvalent"], + "mobile": ["mobile","移动端","ios","android","swift","kotlin","react native","flutter","xamarin","application mobile"], + "devops": ["devops","sre","运维","infrastructure","kubernetes","k8s","docker","ci/cd","pipeline", + "terraform","ansible","aws","gcp","azure","cloud","nuage"], + "ml": ["machine learning","deep learning","机器学习","深度学习","algorithme","ai engineer", + "pytorch","tensorflow","nlp","cv","computer vision","llm","大模型","ia","intelligence artificielle"], + "data": ["data engineer","data analyst","数据工程","数据分析","etl","spark","flink", + "hadoop","kafka","数仓","数据仓库","bi","tableau","airflow","données","analyse"], + "security": ["security","安全","penetration","渗透","ctf","soc","siem","sécurité","cybersécurité"], + "embedded": ["embedded","嵌入式","firmware","rtos","stm32","fpga","驱动","内核","embarqué","temps réel"], +} + +SPECIFIC_QUESTIONS: dict[str, list] = { + "frontend": [ + {"id":"s1","question":"Frameworks front-end requis (plusieurs choix possibles)", + "type":"multiple_choice","options":["React","Vue.js","Angular","Svelte","JS/TS natif"],"weight":5}, + {"id":"s2","question":"Niveau TypeScript exigé", + "type":"single_choice","options":["Maîtrise obligatoire","Expérience souhaitée","Atout","Non requis"],"weight":4}, + {"id":"s3","question":"Solutions CSS / styles (plusieurs choix possibles)", + "type":"multiple_choice","options":["Tailwind CSS","CSS Modules","Styled-components","SCSS/LESS","Libre"],"weight":3}, + {"id":"s4","question":"Expérience en optimisation des performances (Core Web Vitals, etc.)", + "type":"single_choice","options":["Obligatoire","Atout","Non requis"],"weight":4}, + {"id":"s5","question":"Expérience en outillage front-end (build tools, bundlers)", + "type":"single_choice","options":["Obligatoire","Atout","Non requis"],"weight":3}, + {"id":"s6","question":"Participation à un Design System ou bibliothèque de composants", + "type":"single_choice","options":["Obligatoire","Atout","Non requis"],"weight":3}, + {"id":"s7","question":"Expérience SSR / SSG (Next.js, Nuxt.js, etc.)", + "type":"single_choice","options":["Obligatoire","Atout","Non requis"],"weight":3}, + ], + "backend": [ + {"id":"s1","question":"Langages de programmation principaux (plusieurs choix possibles)", + "type":"multiple_choice","options":["Python","Java","Go","Node.js","C++","Rust","PHP","Ruby"],"weight":5}, + {"id":"s2","question":"Bases de données requises (plusieurs choix possibles)", + "type":"multiple_choice","options":["PostgreSQL","MySQL","MongoDB","Redis","Elasticsearch","ClickHouse","Libre"],"weight":4}, + {"id":"s3","question":"Expérience microservices / systèmes distribués", + "type":"single_choice","options":["Obligatoire (3+ ans)","Expérience souhaitée","Atout","Non requis"],"weight":4}, + {"id":"s4","question":"Style d'API requis (plusieurs choix possibles)", + "type":"multiple_choice","options":["RESTful","GraphQL","gRPC","WebSocket","Libre"],"weight":3}, + {"id":"s5","question":"Expérience en messagerie (plusieurs choix possibles)", + "type":"multiple_choice","options":["Kafka","RabbitMQ","RocketMQ","Redis Pub/Sub","Non requis"],"weight":3}, + {"id":"s6","question":"Expérience cloud", + "type":"single_choice","options":["Cloud-native obligatoire","Usage de base suffisant","Atout","Non requis"],"weight":3}, + {"id":"s7","question":"Conception de systèmes haute disponibilité / haute charge", + "type":"single_choice","options":["Obligatoire (million RPS)","Échelle moyenne","Atout","Non requis"],"weight":5}, + ], + "fullstack": [ + {"id":"s1","question":"Préférence stack front-end (plusieurs choix possibles)", + "type":"multiple_choice","options":["React","Vue.js","Angular","Next.js","Nuxt","Libre"],"weight":4}, + {"id":"s2","question":"Préférence stack back-end (plusieurs choix possibles)", + "type":"multiple_choice","options":["Node.js","Python","Java","Go","PHP","Libre"],"weight":4}, + {"id":"s3","question":"Répartition front / back du poste", + "type":"single_choice","options":["Front dominant (70/30)","Équilibré (50/50)","Back dominant (30/70)","Flexible"],"weight":3}, + {"id":"s4","question":"Expérience de développement produit complet en autonomie", + "type":"single_choice","options":["Obligatoire","Atout","Non requis"],"weight":4}, + {"id":"s5","question":"Bases de données (plusieurs choix possibles)", + "type":"multiple_choice","options":["PostgreSQL","MySQL","MongoDB","Redis","Libre"],"weight":3}, + {"id":"s6","question":"Expérience DevOps / déploiement (Docker, CI/CD)", + "type":"single_choice","options":["Obligatoire","Atout","Non requis"],"weight":3}, + ], + "mobile": [ + {"id":"s1","question":"Plateformes cibles (plusieurs choix possibles)", + "type":"multiple_choice","options":["iOS (Swift/ObjC)","Android (Kotlin/Java)","React Native","Flutter","HarmonyOS"],"weight":5}, + {"id":"s2","question":"Expérience de publication App Store / Google Play", + "type":"single_choice","options":["Obligatoire","Atout","Non requis"],"weight":4}, + {"id":"s3","question":"Optimisation des performances (mémoire, rendu, batterie)", + "type":"single_choice","options":["Obligatoire","Atout","Non requis"],"weight":4}, + {"id":"s4","question":"Développement de modules natifs / hybride", + "type":"single_choice","options":["Obligatoire","Atout","Non requis"],"weight":3}, + {"id":"s5","question":"CI/CD et automatisation des builds / releases", + "type":"single_choice","options":["Obligatoire","Atout","Non requis"],"weight":3}, + ], + "devops": [ + {"id":"s1","question":"Plateformes cloud utilisées (plusieurs choix possibles)", + "type":"multiple_choice","options":["AWS","GCP","Azure","Alibaba Cloud","Tencent Cloud","OVH"],"weight":5}, + {"id":"s2","question":"Expérience Kubernetes", + "type":"single_choice","options":["Expert (production)","Expérience en utilisation","Notions suffisantes","Non requis"],"weight":5}, + {"id":"s3","question":"Outils IaC (plusieurs choix possibles)", + "type":"multiple_choice","options":["Terraform","Ansible","Pulumi","CloudFormation","Non requis"],"weight":4}, + {"id":"s4","question":"Monitoring / alerting (plusieurs choix possibles)", + "type":"multiple_choice","options":["Prometheus/Grafana","ELK Stack","Datadog","Libre"],"weight":3}, + {"id":"s5","question":"Gestion de clusters à grande échelle (100+ nœuds)", + "type":"single_choice","options":["Obligatoire","Atout","Non requis"],"weight":4}, + {"id":"s6","question":"Compétences en développement / scripting", + "type":"single_choice","options":["Dev requis (Python/Go)","Scripts shell suffisants","Non requis"],"weight":3}, + ], + "ml": [ + {"id":"s1","question":"Frameworks ML requis (plusieurs choix possibles)", + "type":"multiple_choice","options":["PyTorch","TensorFlow","JAX","scikit-learn","XGBoost","Libre"],"weight":5}, + {"id":"s2","question":"Profil recherche", + "type":"single_choice","options":["Publications (NeurIPS/ICML/ACL…)","Publications souhaitées","Ingénierie avant tout","Non requis"],"weight":4}, + {"id":"s3","question":"Expérience en déploiement de modèles", + "type":"single_choice","options":["Obligatoire (TorchServe/Triton/ONNX)","Atout","Non requis"],"weight":4}, + {"id":"s4","question":"Expérience LLM / grands modèles", + "type":"single_choice","options":["Obligatoire (Fine-tuning/RAG/Agent)","Atout","Non requis"],"weight":4}, + {"id":"s5","question":"Volume de données traité", + "type":"single_choice","options":["Échelle TB+","Échelle GB suffisant","Non requis"],"weight":3}, + {"id":"s6","question":"Entraînement sur clusters GPU", + "type":"single_choice","options":["Obligatoire (distribué)","Multi-GPU mono-machine","Atout","Non requis"],"weight":4}, + ], + "data": [ + {"id":"s1","question":"Stack data requise (plusieurs choix possibles)", + "type":"multiple_choice","options":["Spark","Flink","Hive","Kafka","Airflow","dbt","Libre"],"weight":5}, + {"id":"s2","question":"Entrepôts de données (plusieurs choix possibles)", + "type":"multiple_choice","options":["ClickHouse","BigQuery","Redshift","Snowflake","Hudi/Iceberg","Libre"],"weight":4}, + {"id":"s3","question":"Volumétrie des données", + "type":"single_choice","options":["Échelle PB","Échelle TB","Échelle GB","Libre"],"weight":4}, + {"id":"s4","question":"Modélisation data warehouse (ODS/DW/DM)", + "type":"single_choice","options":["Obligatoire","Atout","Non requis"],"weight":4}, + {"id":"s5","question":"Outils BI (plusieurs choix possibles)", + "type":"multiple_choice","options":["Tableau","Power BI","Superset","Metabase","Non requis"],"weight":3}, + {"id":"s6","question":"Langages requis (plusieurs choix possibles)", + "type":"multiple_choice","options":["Python","SQL (expert)","Scala","Java","Libre"],"weight":4}, + ], + "security": [ + {"id":"s1","question":"Domaines de sécurité (plusieurs choix possibles)", + "type":"multiple_choice","options":["Tests d'intrusion","Audit de code","Développement sécurisé (SDL)","SOC/Réponse incident","Sécurité cloud","Libre"],"weight":5}, + {"id":"s2","question":"Découverte de vulnérabilités / CVE publiés", + "type":"single_choice","options":["Obligatoire (CVE publiés préférés)","Atout","Non requis"],"weight":4}, + {"id":"s3","question":"Outils de sécurité (plusieurs choix possibles)", + "type":"multiple_choice","options":["Burp Suite","Metasploit","Nmap","Wireshark","Outils maison"],"weight":3}, + {"id":"s4","question":"Conformité réglementaire (ISO 27001, RGPD, etc.)", + "type":"single_choice","options":["Obligatoire","Atout","Non requis"],"weight":3}, + ], + "embedded": [ + {"id":"s1","question":"Domaines embarqués (plusieurs choix possibles)", + "type":"multiple_choice","options":["Développement drivers","RTOS","BSP","FPGA","Firmware"],"weight":5}, + {"id":"s2","question":"Plateformes matérielles (plusieurs choix possibles)", + "type":"multiple_choice","options":["ARM Cortex-M","ARM Cortex-A","RISC-V","x86","DSP","Libre"],"weight":4}, + {"id":"s3","question":"Protocoles de communication (plusieurs choix possibles)", + "type":"multiple_choice","options":["SPI/I2C/UART","CAN","Ethernet","BLE/WiFi","Libre"],"weight":3}, + {"id":"s4","question":"Développement noyau Linux", + "type":"single_choice","options":["Obligatoire","Atout","Non requis"],"weight":4}, + ], + "general": [ + {"id":"s1","question":"Niveau de spécialisation technique", + "type":"single_choice","options":["Expert (spécialité pointue)","Généraliste (large spectre)","Manager technique","Libre"],"weight":4}, + {"id":"s2","question":"Taille d'équipe / projet préférée", + "type":"single_choice","options":["Grande entreprise (100+)","Équipe moyenne (20-100)","Startup / petite équipe (<20)","Libre"],"weight":3}, + {"id":"s3","question":"Contributions open-source", + "type":"single_choice","options":["Obligatoire (100+ stars)","Atout","Non requis"],"weight":3}, + {"id":"s4","question":"Expérience management d'équipe", + "type":"single_choice","options":["Obligatoire (5+ personnes)","Petite équipe suffisant","Non requis"],"weight":4}, + {"id":"s5","question":"Documentation technique / conférences", + "type":"single_choice","options":["Obligatoire (blog/talks)","Atout","Non requis"],"weight":2}, + {"id":"s6","question":"Collaboration cross-équipes", + "type":"single_choice","options":["Très important","Important","Neutre","Peu important"],"weight":3}, + ], +} + +# ── Nouvelle question stabilité (ajoutée aux questions générales) ────────────── +STABILITY_QUESTION = { + "id": "g_stab", + "question": "Préférence en matière de stabilité des employés", + "type": "single_choice", + "options": [ + "Stabilité forte (durée moyenne ≥ 3 ans par poste)", + "Stabilité modérée (1–3 ans par poste)", + "Mobilité acceptée (CDD, freelance, missions courtes)", + "Sans critère de stabilité", + ], + "weight": 3, +} + +GENERAL_QUESTIONS = [ + {"id":"g1","question":"Mode de travail souhaité", + "type":"single_choice","options":["100% télétravail","Hybride (partiel télétravail)","Présentiel","Flexible"],"weight":3}, + {"id":"g2","question":"Niveau d'études minimum", + "type":"single_choice","options":["Sans critère","Bac+2","Licence (grandes écoles préférées)","Licence","Master","Doctorat"],"weight":3}, + {"id":"g3","question":"Années d'expérience minimum", + "type":"single_choice","options":["Sans critère","1 an et +","3 ans et +","5 ans et +","8 ans et +","10 ans et +"],"weight":5}, + {"id":"g4","question":"Langues requises (plusieurs choix possibles)", + "type":"multiple_choice","options":["Français (langue maternelle)","Anglais (lu/écrit)","Anglais (courant oral)","Espagnol","Autre"],"weight":3}, + {"id":"g5","question":"Fourchette de salaire mensuel brut (indicatif)", + "type":"single_choice", + "options":["< 2 500 €","2 500–4 000 €","4 000–6 000 €","6 000–8 500 €","8 500 € +","À négocier"],"weight":1}, + STABILITY_QUESTION, # ← nouvelle question v2 +] + + +# ─── Générateur de questionnaire ────────────────────────────────────────────── + +def _detect_category(title: str, description: str) -> str: + text = (title + " " + description).lower() + scores: dict[str, int] = {} + for cat, keywords in CATEGORY_SIGNALS.items(): + score = sum(1 for kw in keywords if kw in text) + if score: + scores[cat] = score + if not scores: + return "general" + return max(scores, key=scores.get) + + +def _extract_mentioned_techs(description: str) -> list[str]: + ALL_TECHS = [ + "React","Vue","Angular","Next.js","Nuxt","Svelte","TypeScript","JavaScript", + "Python","Java","Go","Golang","Rust","C++","C#","PHP","Ruby","Swift","Kotlin", + "Django","Flask","FastAPI","Spring","Rails","Express","NestJS","Laravel", + "PostgreSQL","MySQL","MongoDB","Redis","Elasticsearch","ClickHouse","SQLite", + "AWS","GCP","Azure","Kubernetes","Docker","Terraform","Ansible","CI/CD","GitHub Actions", + "Kafka","RabbitMQ","Spark","Flink","Airflow","dbt","Hadoop", + "PyTorch","TensorFlow","scikit-learn","LLM","RAG","GPT","BERT", + "GraphQL","gRPC","REST","WebSocket","Prometheus","Grafana","ELK", + "Flutter","React Native","iOS","Android","SwiftUI", + ] + found = [] + desc_lower = description.lower() + for tech in ALL_TECHS: + if tech.lower() in desc_lower and tech not in found: + found.append(tech) + return found[:8] + + +def _generate_questionnaire(job_title: str, job_description: str) -> dict: + category = _detect_category(job_title, job_description) + specific_qs = SPECIFIC_QUESTIONS.get(category, SPECIFIC_QUESTIONS["general"]) + + mentioned = _extract_mentioned_techs(job_description) + extra_qs = [] + if mentioned: + extra_qs.append({ + "id": "s_dyn", + "question": "Parmi les technologies mentionnées dans la fiche de poste, lesquelles sont indispensables ? (plusieurs choix possibles)", + "type": "multiple_choice", + "options": mentioned + ["Toutes sont des atouts"], + "weight": 5, + }) + + return { + "general": GENERAL_QUESTIONS, + "specific": extra_qs + specific_qs, + "_category": category, + "_mentioned_techs": mentioned, + } + + +# ─── Stabilité des candidats ────────────────────────────────────────────────── + +def _average_tenure_years(profile: dict) -> float: + """Calcule la durée moyenne (en années) par poste dans le profil.""" + total_years = float(profile.get("experiences_duration") or 0) + n_exp = len(profile.get("experiences", [])) + if total_years <= 0 or n_exp == 0: + return 0.0 + return round(total_years / n_exp, 1) + + +# ─── Interprétation des réponses ────────────────────────────────────────────── + +def _interpret_answers(questionnaire: dict, answers: dict, user_weights: dict = None) -> dict: + """user_weights : { qid: int(1-5) } — overrides question's default weight.""" + q_map: dict[str, dict] = {} + for q in questionnaire.get("general", []) + questionnaire.get("specific", []): + q_map[q["id"]] = q + + required_skills: list[str] = [] + preferred_skills: list[str] = [] + min_years = 0 + max_years = 0 + education_level = "none" + work_mode = "any" + languages: list[str] = [] + stability_pref = "none" # "high" | "moderate" | "flexible" | "none" + + def effective_weight(qid: str, default: int) -> int: + if user_weights and qid in user_weights: + return int(user_weights[qid]) + return default + + for qid, answer in answers.items(): + q = q_map.get(qid, {}) + q_text = q.get("question", "").lower() + weight = effective_weight(qid, q.get("weight", 3)) + ans_list = answer if isinstance(answer, list) else [answer] if answer else [] + + # Stabilité + if "stabilité" in q_text: + for a in ans_list: + a_lower = a.lower() + if "≥ 3" in a or "forte" in a_lower: + stability_pref = "high" + elif "1–3" in a or "modérée" in a_lower: + stability_pref = "moderate" + elif "mobilité" in a_lower or "acceptée" in a_lower: + stability_pref = "flexible" + continue + + # Mode de travail + if "mode de travail" in q_text or "télétravail" in q_text: + mode_map = {"100% télétravail": "remote", "hybride": "hybrid", "présentiel": "onsite"} + for k, v in mode_map.items(): + if any(k in a.lower() for a in ans_list): + work_mode = v + break + + # Niveau d'études + elif "études" in q_text or "niveau" in q_text: + edu_map = {"doctorat": "phd", "master": "master", "licence": "bachelor", "bac+2": "none"} + for k, v in edu_map.items(): + if any(k in a.lower() for a in ans_list): + education_level = v + break + + # Années d'expérience + elif "expérience" in q_text and ("ans" in str(ans_list).lower() or "an" in str(ans_list).lower()): + for a in ans_list: + m = re.search(r"(\d+)", str(a)) + if m: + min_years = max(min_years, int(m.group(1))) + + # Langues + elif "langues" in q_text: + languages = [a for a in ans_list if "autre" not in a.lower() and "français" not in a.lower()] + + # Technologies fiche de poste + elif "fiche de poste" in q_text or "mentionnées" in q_text: + for a in ans_list: + if "atout" in a.lower(): + continue + if a and a not in required_skills: + required_skills.append(a) + + # Compétences / outils — seuil weight dynamique + elif any(kw in q_text for kw in + ["framework","stack","langage","outil","plateforme","domaine","base de données", + "messagerie","protocole","matérielle","entrepôt"]): + clean = [a for a in ans_list if a and "libre" not in a.lower() and "non requis" not in a.lower()] + if weight >= 4: + required_skills.extend(c for c in clean if c not in required_skills) + elif weight >= 2: + preferred_skills.extend(c for c in clean if c not in preferred_skills) + + # Questions binaires + elif any(kw in q_text for kw in ["obligatoire","expérience","participation","niveau","conception"]): + for a in ans_list: + if "obligatoire" in a.lower() or "maîtrise obligatoire" in a.lower(): + concept = _extract_concept_fr(q_text) + if concept and concept not in required_skills: + required_skills.append(concept) + elif "atout" in a.lower(): + concept = _extract_concept_fr(q_text) + if concept and concept not in preferred_skills: + preferred_skills.append(concept) + + key_parts = [] + if required_skills: + key_parts.append(", ".join(required_skills[:4])) + if min_years > 0: + key_parts.append(f"{min_years}+ ans d'expérience") + if education_level != "none": + key_parts.append({"bachelor":"Licence","master":"Master","phd":"Doctorat"}.get(education_level,"") + " requis") + + return { + "required_skills": required_skills, + "preferred_skills": preferred_skills, + "min_experience_years": min_years, + "max_experience_years": max_years, + "education_level": education_level, + "work_mode": work_mode, + "languages": languages, + "stability_pref": stability_pref, + "key_requirements_summary": ", ".join(key_parts) if key_parts else "Profil polyvalent", + } + + +def _extract_concept_fr(q_text: str) -> str: + CONCEPT_MAP = { + "typescript": "TypeScript", "performance": "Optimisation des performances", + "outillage": "Build tools", "ssr": "SSR/SSG", "design system": "Design System", + "microservice": "Microservices", "distribué": "Systèmes distribués", "haute disponibilité": "Haute disponibilité", + "déploiement de modèles": "Déploiement ML", "llm": "LLM/Grands modèles", "gpu": "Entraînement GPU", + "open-source": "Open-source", "management": "Management", "kubernetes": "Kubernetes", + "noyau linux": "Noyau Linux", "iac": "IaC", + } + for kw, label in CONCEPT_MAP.items(): + if kw in q_text: + return label + return "" + + +# ─── Scoring des candidats ──────────────────────────────────────────────────── + +def _calculate_experience_years(profile: dict) -> float: + return round(float(profile.get("experiences_duration") or 0), 1) + + +def _get_skills(profile: dict) -> set: + return {s.get("name", "").lower() for s in profile.get("skills", []) if s.get("name")} + + +def _edu_level(profile: dict) -> str: + for edu in profile.get("educations", []): + t = (edu.get("title") or "").lower() + if any(k in t for k in ["phd","doctorate","doctorat"]): return "phd" + if any(k in t for k in ["master","msc","mba","mastère"]): return "master" + if any(k in t for k in ["bachelor","bsc","undergraduate","licence","ingénieur"]): return "bachelor" + return "none" + + +_EDU_RANK = {"none": 0, "bachelor": 1, "master": 2, "phd": 3} +_EDU_LABEL_FR = {"phd": "Doctorat", "master": "Master", "bachelor": "Licence", "none": "Non renseigné"} + + +def _score_candidates(profiles: list, reqs: dict, user_weights: dict = None) -> list: + required = {s.lower() for s in reqs.get("required_skills", [])} + preferred = {s.lower() for s in reqs.get("preferred_skills", [])} + min_years = float(reqs.get("min_experience_years") or 0) + max_years = float(reqs.get("max_experience_years") or 99) + edu_req = _EDU_RANK.get(reqs.get("education_level", "none"), 0) + stability_pref = reqs.get("stability_pref", "none") + + # Dimension multipliers from weight sliders (1–5 scale → 0.2–1.0) + uw = user_weights or {} + w_exp = uw.get("g3", 5) / 5.0 + w_edu = uw.get("g2", 3) / 5.0 + s_keys = [k for k in uw if k.startswith("s")] + w_skill = (sum(uw[k] for k in s_keys) / len(s_keys) / 5.0) if s_keys else 1.0 + w_stab = uw.get("g_stab", 3) / 5.0 + + # Max achievable raw score (used for normalization) + max_possible = 50 * w_skill + 30 * w_exp + 20 * w_edu + 8 * w_stab + + scored = [] + for p in profiles: + skills = _get_skills(p) + years = _calculate_experience_years(p) + edu = _edu_level(p) + avg_tenure = _average_tenure_years(p) + info = p.get("info", {}) + name = (info.get("full_name") + or f"{info.get('first_name','')} {info.get('last_name','')}".strip() + or p.get("key", "")[:12]) + + raw = 0.0 + + # Compétences — 50 pts × w_skill + skill_pts = 0.0 + if required: + matched_req = required & skills + skill_pts += (len(matched_req) / len(required)) * 50 + else: + skill_pts += 30 + if preferred: + pref_hit = preferred & skills + skill_pts += (len(pref_hit) / len(preferred)) * 20 + raw += skill_pts * w_skill + + # Expérience — 30 pts × w_exp + exp_pts = 0.0 + if min_years > 0: + if years >= min_years: exp_pts += 30 + elif years >= min_years * 0.75: exp_pts += 15 + elif years >= min_years * 0.5: exp_pts += 5 + else: + exp_pts += 30 + if max_years < 99 and years > max_years + 5: + exp_pts -= 8 + raw += exp_pts * w_exp + + # Formation — 20 pts × w_edu + edu_pts = 0.0 + cand_edu_rank = _EDU_RANK.get(edu, 0) + if cand_edu_rank >= edu_req: edu_pts += 20 + elif cand_edu_rank == edu_req - 1: edu_pts += 10 + raw += edu_pts * w_edu + + # Stabilité — ±8 pts × w_stab + stability_note = "" + stab_pts = 0.0 + if stability_pref == "high": + if avg_tenure >= 3.0: + stab_pts = 8 + stability_note = f"Bonne stabilité ({avg_tenure:.1f} ans/poste en moy.)" + elif avg_tenure > 0: + stab_pts = -5 + stability_note = f"Mobilité fréquente ({avg_tenure:.1f} ans/poste en moy.)" + elif stability_pref == "moderate": + if 1.0 <= avg_tenure < 4.0: + stab_pts = 4 + stability_note = f"Stabilité modérée ({avg_tenure:.1f} ans/poste en moy.)" + elif avg_tenure >= 4.0: + stability_note = f"Profil stable ({avg_tenure:.1f} ans/poste en moy.)" + elif stability_pref == "flexible": + if avg_tenure < 1.5: + stab_pts = 4 + stability_note = f"Profil mobile ({avg_tenure:.1f} ans/poste en moy.)" + raw += stab_pts * w_stab + + matched_req_list = sorted(required & skills) + missing_req_list = sorted(required - skills) + bonus_list = sorted((preferred & skills) - required) + + # Normalize to 0–100 with decimal precision + if max_possible > 0: + final_score = round(min(raw / max_possible * 100, 100), 2) + else: + final_score = 0.0 + + # ── Mention ────────────────────────────────────────────── + if final_score >= 85: grade = "A" + elif final_score >= 70: grade = "B" + elif final_score >= 50: grade = "C" + else: grade = "D" + + # ── Justification ───────────────────────────────────────── + parts = [] + if required: + ratio = len(matched_req_list) / len(required) + if ratio == 1.0: + parts.append(f"Toutes les {len(required)} compétences requises ({', '.join(matched_req_list[:4])})") + elif ratio >= 0.6: + parts.append(f"{len(matched_req_list)}/{len(required)} compétences requises" + + (f" — manque : {', '.join(missing_req_list[:2])}" if missing_req_list else "")) + else: + parts.append(f"Seulement {len(matched_req_list)}/{len(required)} requises — manque {', '.join(missing_req_list[:3])}") + if min_years > 0: + label = f"{years:.1f} ans exp. (requis : {min_years:.0f}+)" if years >= min_years else f"{years:.1f} ans exp. (< {min_years:.0f} requis)" + parts.append(label) + elif years > 0: + parts.append(f"{years:.1f} ans d'expérience") + edu_label = _EDU_LABEL_FR.get(edu, "") + if edu_label and edu_label != "Non renseigné": + parts.append(edu_label) + if stability_note: + parts.append(stability_note) + if bonus_list: + parts.append(f"Bonus : {', '.join(bonus_list[:3])}") + reasoning = " ; ".join(parts) if parts else "Évaluation sur profil global" + + scored.append({ + "key": p.get("key", ""), + "name": name, + "email": info.get("email", ""), + "summary": (info.get("summary") or "")[:300], + "skills": sorted(skills), + "experience_years": years, + "avg_tenure": avg_tenure, + "education": edu, + "matched_skills": matched_req_list, + "missing_skills": missing_req_list, + "bonus_skills": bonus_list, + "score": final_score, + "grade": grade, + "reasoning": reasoning, + }) + + scored.sort(key=lambda x: x["score"], reverse=True) + return scored + + +def _score_with_hrflow(profiles: list, predictions: list, reqs: dict, user_weights: dict = None) -> list: + """ + Combine HrFlow AI predictions (Leo's algorithm_key) avec les ajustements + du questionnaire pondérés par les sliders utilisateur. + Score final = score_IA_HrFlow (0-100) + delta questionnaire (±30 pts max). + """ + required = {s.lower() for s in reqs.get("required_skills", [])} + preferred = {s.lower() for s in reqs.get("preferred_skills", [])} + min_years = float(reqs.get("min_experience_years") or 0) + max_years = float(reqs.get("max_experience_years") or 99) + edu_req = _EDU_RANK.get(reqs.get("education_level", "none"), 0) + stability_pref = reqs.get("stability_pref", "none") + + uw = user_weights or {} + w_exp = uw.get("g3", 5) / 5.0 + w_edu = uw.get("g2", 3) / 5.0 + s_keys = [k for k in uw if k.startswith("s")] + w_skill = (sum(uw[k] for k in s_keys) / len(s_keys) / 5.0) if s_keys else 1.0 + w_stab = uw.get("g_stab", 3) / 5.0 + + scored = [] + for i, p in enumerate(profiles): + # Score de base HrFlow IA (predictions[i][1] = probabilité 0-1) + ai_score = 0.0 + if i < len(predictions) and len(predictions[i]) > 1: + ai_score = round(predictions[i][1] * 100, 2) + + skills = _get_skills(p) + years = _calculate_experience_years(p) + edu = _edu_level(p) + avg_tenure = _average_tenure_years(p) + info = p.get("info", {}) + name = (info.get("full_name") + or f"{info.get('first_name','')} {info.get('last_name','')}".strip() + or p.get("key", "")[:12]) + + delta = 0.0 + + # Compétences : ±15 pts pondérés + matched_req_list = sorted(required & skills) if required else [] + missing_req_list = sorted(required - skills) if required else [] + bonus_list = sorted((preferred & skills) - required) if preferred else [] + if required: + skill_ratio = len(matched_req_list) / len(required) + delta += (skill_ratio - 0.5) * 20 * w_skill + + # Expérience : ±10 pts pondérés + if min_years > 0: + if years >= min_years: delta += 8 * w_exp + elif years >= min_years * 0.75: delta += 2 * w_exp + else: delta -= 8 * w_exp + if max_years < 99 and years > max_years + 5: + delta -= 5 * w_exp + + # Formation : ±8 pts pondérés + cand_edu_rank = _EDU_RANK.get(edu, 0) + if cand_edu_rank >= edu_req: delta += 5 * w_edu + elif cand_edu_rank < edu_req - 1: delta -= 8 * w_edu + + # Stabilité : ±5 pts pondérés + stability_note = "" + if stability_pref == "high": + if avg_tenure >= 3.0: + delta += 5 * w_stab + stability_note = f"Bonne stabilité ({avg_tenure:.1f} ans/poste)" + elif avg_tenure > 0: + delta -= 4 * w_stab + stability_note = f"Mobilité fréquente ({avg_tenure:.1f} ans/poste)" + elif stability_pref == "moderate": + if 1.0 <= avg_tenure < 4.0: + delta += 2 * w_stab + stability_note = f"Stabilité modérée ({avg_tenure:.1f} ans/poste)" + elif stability_pref == "flexible": + if avg_tenure < 1.5: + delta += 2 * w_stab + stability_note = f"Profil mobile ({avg_tenure:.1f} ans/poste)" + + final_score = round(min(max(ai_score + delta, 0), 100), 2) + + if final_score >= 85: grade = "A" + elif final_score >= 70: grade = "B" + elif final_score >= 50: grade = "C" + else: grade = "D" + + parts = [f"Score HrFlow IA : {ai_score:.1f}/100"] + if required: + ratio = len(matched_req_list) / len(required) + if ratio == 1.0: + parts.append(f"Toutes les {len(required)} compétences requises ({', '.join(matched_req_list[:3])})") + elif ratio >= 0.5: + parts.append(f"{len(matched_req_list)}/{len(required)} compétences" + + (f" — manque : {', '.join(missing_req_list[:2])}" if missing_req_list else "")) + else: + parts.append(f"Seulement {len(matched_req_list)}/{len(required)} requises — manque {', '.join(missing_req_list[:2])}") + if min_years > 0: + parts.append(f"{years:.1f} ans exp. (requis : {min_years:.0f}+)" if years >= min_years + else f"{years:.1f} ans exp. (< {min_years:.0f} requis)") + elif years > 0: + parts.append(f"{years:.1f} ans d'expérience") + edu_label = _EDU_LABEL_FR.get(edu, "") + if edu_label and edu_label != "Non renseigné": + parts.append(edu_label) + if stability_note: + parts.append(stability_note) + if bonus_list: + parts.append(f"Bonus : {', '.join(bonus_list[:3])}") + reasoning = " ; ".join(parts) if parts else "Évaluation HrFlow IA" + + scored.append({ + "key": p.get("key", ""), + "name": name, + "email": info.get("email", ""), + "summary": (info.get("summary") or "")[:300], + "skills": sorted(skills), + "experience_years": years, + "avg_tenure": avg_tenure, + "education": edu, + "matched_skills": matched_req_list, + "missing_skills": missing_req_list, + "bonus_skills": bonus_list, + "score": final_score, + "ai_base_score": ai_score, + "grade": grade, + "reasoning": reasoning, + }) + + scored.sort(key=lambda x: x["score"], reverse=True) + return scored + + +# ─── Analyse du vivier ──────────────────────────────────────────────────────── + +def _analyse_candidates(candidates: list) -> dict: + if len(candidates) < 3: + return {"common_skills": [], "differentiators": [], "experience_range": "", "score_range": "", "candidate_count": len(candidates)} + + top = candidates[:min(20, len(candidates))] + n = len(top) + threshold = 0.65 + + all_skills_flat = [s for c in top for s in c["skills"]] + skill_counts = Counter(all_skills_flat) + + common_skills = [s for s, cnt in skill_counts.most_common(30) if cnt >= n * threshold] + differentiators = [s for s, cnt in skill_counts.most_common(50) if n * 0.15 <= cnt < n * threshold] + + years_list = [c["experience_years"] for c in top] + scores = [c["score"] for c in top] + + return { + "candidate_count": n, + "common_skills": common_skills[:12], + "differentiators": differentiators[:12], + "experience_range": f"{min(years_list):.0f}–{max(years_list):.0f} ans" if years_list else "", + "score_range": f"{min(scores):.0f}–{max(scores):.0f}" if scores else "", + "education_distribution": dict(Counter(c["education"] for c in top)), + } + + +# ─── Affinement ─────────────────────────────────────────────────────────────── + +def _apply_refinement(candidates: list, refinement_text: str, original_reqs: dict, user_weights: dict = None) -> list: + text_lower = refinement_text.lower() + updated = dict(original_reqs) + + ALL_TECHS = [ + "react","vue","angular","typescript","javascript","python","java","go","rust", + "kubernetes","docker","aws","gcp","azure","kafka","spark","pytorch","tensorflow", + "llm","graphql","grpc","redis","mongodb","postgresql","mysql", + "flutter","swift","kotlin","next.js","nuxt","terraform","ansible", + "elasticsearch","clickhouse","airflow","dbt","scikit-learn","rag", + ] + extra_required = [t for t in ALL_TECHS if t in text_lower] + if extra_required: + existing = [s.lower() for s in updated.get("required_skills", [])] + for t in extra_required: + if t not in existing: + updated.setdefault("required_skills", []).append(t) + + m = re.search(r"(\d+)\s*(?:ans?|années?)", refinement_text, re.IGNORECASE) + if m: + updated["min_experience_years"] = max(updated.get("min_experience_years", 0), int(m.group(1))) + + if any(k in text_lower for k in ["doctorat","phd"]): + updated["education_level"] = "phd" + elif any(k in text_lower for k in ["master","mastère","m2"]): + if _EDU_RANK.get(updated.get("education_level","none"),0) < 2: + updated["education_level"] = "master" + + if "stable" in text_lower or "stabilité" in text_lower: + updated["stability_pref"] = "high" + elif "mobile" in text_lower or "freelance" in text_lower: + updated["stability_pref"] = "flexible" + + fake_profiles = [] + for c in candidates: + fake_profiles.append({ + "key": c["key"], + "info": {"full_name": c["name"], "email": c["email"], "summary": c["summary"]}, + "skills": [{"name": s} for s in c["skills"]], + "experiences": [{}] * max(1, round(c["experience_years"] / max(c.get("avg_tenure", 1), 0.1))), + "experiences_duration": c["experience_years"], + "educations": _fake_edu(c["education"]), + }) + return _score_candidates(fake_profiles, updated, user_weights) + + +def _fake_exp_tenure(total_years: float, avg_tenure: float) -> list: + if total_years <= 0: + return [] + from datetime import timedelta + tenure = avg_tenure if avg_tenure > 0 else total_years + n_jobs = max(1, round(total_years / tenure)) + exps = [] + end = datetime.now() + for _ in range(n_jobs): + start = end - timedelta(days=int(tenure * 365)) + exps.append({"title":"Engineer","company":"Company", + "date_start": start.isoformat(),"date_end": end.isoformat()}) + end = start + return exps + + +def _fake_edu(level: str) -> list: + titles = {"bachelor":"Licence","master":"Master","phd":"Doctorat","none":""} + t = titles.get(level, "") + return [{"title": t}] if t else [] + + +# ─── API Endpoints ──────────────────────────────────────────────────────────── + +@app.get("/") +def root(): + return FileResponse("static/index_openhr.html") + + +@app.post("/api/start") +def api_start(req: StartRequest): + if not req.job_title.strip(): + raise HTTPException(400, "L'intitulé du poste est requis") + questionnaire = _generate_questionnaire(req.job_title, req.job_description) + session_id = str(uuid.uuid4()) + sessions[session_id] = { + "job_title": req.job_title, + "job_description": req.job_description, + "questionnaire": questionnaire, + "candidates": [], + "requirements": {}, + } + return {"session_id": session_id, "questionnaire": questionnaire} + + +@app.post("/api/submit") +def api_submit(req: SubmitAnswersRequest): + session = sessions.get(req.session_id) + if not session: + raise HTTPException(404, "Session introuvable, veuillez actualiser") + + user_weights = req.weights or {} + reqs = _interpret_answers(session["questionnaire"], req.answers, user_weights) + session["requirements"] = reqs + session["user_weights"] = user_weights + + try: + # Étape 1 : indexer le poste sur le board pour obtenir un job_key + job_key = hrflow.index_job_sync(config.HRFLOW_BOARD_KEY, session["job_title"]) + session["job_key"] = job_key + + # Étape 2 : attendre que HrFlow calcule les scores (approche Leo) + time.sleep(5) + + # Étape 3 : récupérer les profils scorés par l'IA HrFlow (algorithm_key) + scoring_resp = hrflow.score_profiles_with_algo( + job_key=job_key, + board_key=config.HRFLOW_BOARD_KEY, + source_keys=[config.HRFLOW_SOURCE_KEY], + limit=100, + ) + raw_profiles = scoring_resp.get("data", {}).get("profiles", []) + predictions = scoring_resp.get("data", {}).get("predictions", []) + + # Fallback : si le scoring API échoue, revenir à la recherche + if not raw_profiles: + search_resp = hrflow.search_profiles( + source_keys=[config.HRFLOW_SOURCE_KEY], + query=session["job_title"], + page=1, limit=100, + ) + raw_profiles = search_resp.get("data", {}).get("profiles", []) + predictions = [] + except Exception as e: + raise HTTPException(500, f"Erreur lors de la récupération des candidats : {e}") + + if predictions: + scored = _score_with_hrflow(raw_profiles, predictions, reqs, user_weights) + else: + scored = _score_candidates(raw_profiles, reqs, user_weights) + session["candidates"] = scored + + qualified = [c for c in scored if c["score"] >= 35] + if not qualified: + qualified = scored[:10] + + analysis = _analyse_candidates(qualified) + + if len(qualified) > 10: + return { + "status": "refine", + "analysis": analysis, + "requirements_summary": reqs.get("key_requirements_summary", ""), + "sample_candidates": qualified[:6], + "total_qualified": len(qualified), + } + + return { + "status": "done", + "candidates": qualified[:15], + "analysis": analysis, + "requirements_summary": reqs.get("key_requirements_summary", ""), + "total_qualified": len(qualified), + } + + +@app.post("/api/refine") +def api_refine(req: RefineRequest): + session = sessions.get(req.session_id) + if not session: + raise HTTPException(404, "Session introuvable") + + candidates = session.get("candidates", []) + if not candidates: + raise HTTPException(400, "Aucun candidat disponible") + + refined = _apply_refinement(candidates, req.refinement, session["requirements"], session.get("user_weights", {})) + qualified = [c for c in refined if c["score"] >= 35] + if not qualified: + qualified = refined[:10] + + analysis = _analyse_candidates(qualified) + return { + "status": "done", + "candidates": qualified[:15], + "analysis": analysis, + "requirements_summary": req.refinement[:80], + "total_qualified": len(qualified), + } + + +@app.get("/api/health") +def health(): + return {"status": "ok", "service": "AI Agent avec OpenClaw v2", "port": 8002}