Blame for app.py

import os
# Matikan fitur OneDNN dan batasi thread agar PyTorch (dari SentenceTransformers) tidak bentrok dengan TensorFlow
os.environ["TF_ENABLE_ONEDNN_OPTS"] = "0"
os.environ["OMP_NUM_THREADS"] = "1"
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3"

import numpy as np
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
import tensorflow as tf
from sentence_transformers import SentenceTransformer

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

CANONICAL_SKILL_MAP = {

    # ==========================================================================
    # PROGRAMMING LANGUAGES
    # ==========================================================================
    "python": ["python", "py"],
    "javascript": ["javascript", "js", "es6", "es2015", "ecmascript"],
    "typescript": ["typescript", "ts"],
    "java": ["java 8", "java 11", "java spring", "java development", "java developer", "java ee", "java se", "java programming"],
    "kotlin": ["kotlin"],
    "php": ["php"],
    "c#": ["c#", "csharp", ".net", "asp.net", "dotnet"],
    "c++": ["c++ programming", "c++ developer", "c/c++", "cpp", "pemrograman c++"],
    "c": ["c language", "bahasa c", "pemrograman c"],
    "go": ["golang", "go lang", "go programming"],
    "rust": ["rust programming", "rust lang"],
    "ruby": ["ruby on rails", "ruby"],
    "swift": ["swift", "swiftui"],
    "dart": ["dart", "dart language"],
    "scala": ["scala"],
    "r": ["r programming", "r language", "rstudio", "r studio"],
    "cobol": ["cobol"],
    "bash/shell": ["bash", "shell script", "shell scripting", "powershell", "zsh"],
    "matlab": ["matlab"],
    "groovy": ["groovy"],

    # ==========================================================================
    # WEB FRONTEND
    # ==========================================================================
    "html": ["html", "html5"],
    "css": ["css", "css3", "stylesheet", "tailwind", "bootstrap"],
    "react.js": ["react.js", "reactjs", "react js", "react hooks", "redux", "react router", "next.js", "nextjs"],
    "vue.js": ["vue.js", "vuejs", "vue js", "vuex", "nuxt.js", "nuxtjs"],
    "angular": ["angular", "angularjs", "angular.js"],
    "svelte": ["svelte"],
    "jquery": ["jquery"],
    "webpack": ["webpack"],
    "vite": ["vite"],

    # ==========================================================================
    # WEB BACKEND & FRAMEWORKS
    # ==========================================================================
    "node.js": ["node.js", "nodejs", "node js", "express.js", "expressjs", "hapi.js", "fastify", "nest.js", "nestjs"],
    "laravel": ["laravel"],
    "django": ["django"],
    "flask": ["flask"],
    "fastapi": ["fastapi", "fast api"],
    "spring boot": ["spring boot", "spring framework", "spring mvc"],
    "asp.net": ["asp.net", "aspnet", ".net core"],
    "ruby on rails": ["ruby on rails", "rails framework"],
    "codeigniter": ["codeigniter"],
    "symfony": ["symfony"],
    "rest api": ["rest api", "restful api", "restful", "api rest", "http api", "web service", "web services"],
    "graphql": ["graphql", "graph ql"],
    "grpc": ["grpc"],
    "websocket": ["websocket", "web socket", "socket.io"],
    "microservices": ["microservices", "microservice", "micro service"],
    "wordpress": ["wordpress"],
    "magento": ["magento"],
    "shopify": ["shopify"],

    # ==========================================================================
    # MOBILE DEVELOPMENT
    # ==========================================================================
    "android": ["android sdk", "android development", "android studio", "android app", "jetpack compose", "android kotlin"],
    "ios": ["ios development", "xcode", "uikit", "ios app"],
    "flutter": ["flutter"],
    "react native": ["react native"],
    "jetpack": ["jetpack", "navigation component", "livedata", "viewmodel", "room database"],

    # ==========================================================================
    # DATABASES
    # ==========================================================================
    "mysql": ["mysql"],
    "postgresql": ["postgresql", "postgres"],
    "mongodb": ["mongodb", "mongo db"],
    "redis": ["redis"],
    "sqlite": ["sqlite"],
    "oracle db": ["oracle db", "oracle database", "oracle sql"],
    "microsoft sql server": ["sql server", "mssql", "ms sql"],
    "elasticsearch": ["elasticsearch", "elastic search"],
    "cassandra": ["cassandra"],
    "dynamodb": ["dynamodb"],
    "firebase": ["firebase", "firestore", "realtime database"],
    "sql": ["sql query", "sql joins", "stored procedure", "sql optimization", "structured query language", "query sql"],
    "nosql": ["nosql", "no-sql"],

    # ==========================================================================
    # DATA SCIENCE & MACHINE LEARNING
    # ==========================================================================
    "machine learning": ["machine learning", "ml model", "supervised learning", "unsupervised learning", "classification", "regression model", "clustering"],
    "deep learning": ["deep learning", "neural network", "dnn", "cnn", "rnn", "lstm", "transformer model"],
    "tensorflow": ["tensorflow", "tensor flow", "keras", "tf.keras"],
    "pytorch": ["pytorch", "torch"],
    "scikit-learn": ["scikit-learn", "sklearn", "scikit learn"],
    "nlp": ["nlp", "natural language processing", "text mining", "text classification", "sentiment analysis", "named entity recognition", "transformers nlp", "bert", "gpt", "llm", "large language model"],
    "computer vision": ["computer vision", "image recognition", "object detection", "opencv", "image processing", "yolo"],
    "pandas": ["pandas"],
    "numpy": ["numpy", "num py"],
    "data visualization": ["matplotlib", "seaborn", "plotly", "data visualization", "visualisasi", "dashboard"],
    "mlops": ["mlops", "ml ops", "model deployment", "model serving", "model monitoring"],
    "data analysis": ["data analysis", "data analyst", "analisis data", "exploratory data analysis", "eda", "statistical analysis", "data wrangling", "data cleaning", "analytics"],
    "statistics": ["statistics", "statistik", "probability", "hypothesis testing", "a/b testing", "regression analysis"],
    "big data": ["big data", "hadoop", "spark", "apache spark", "hive", "kafka", "databricks", "data lake"],

    # ==========================================================================
    # DATA ENGINEERING & BI
    # ==========================================================================
    "etl": ["etl", "extract transform load", "data pipeline", "data ingestion", "airflow", "luigi"],
    "data warehouse": ["data warehouse", "data warehousing", "snowflake", "redshift", "bigquery", "dbt"],
    "power bi": ["power bi", "powerbi", "dax", "power query"],
    "tableau": ["tableau"],
    "looker": ["looker", "looker studio"],
    "google analytics": ["google analytics", "ga4", "universal analytics"],
    "excel": ["microsoft excel", "ms excel", "excel", "pivot table", "vlookup", "macro excel", "spreadsheet"],
    "business intelligence": ["business intelligence", "reporting dashboard", "kpi dashboard"],

    # ==========================================================================
    # CLOUD & DEVOPS
    # ==========================================================================
    "aws": ["aws", "amazon web services", "ec2", "s3 bucket", "lambda aws", "rds aws", "cloudfront", "eks", "ecs"],
    "google cloud": ["google cloud", "gcp", "bigquery", "vertex ai", "cloud run", "cloud storage", "cloud functions"],
    "microsoft azure": ["azure", "microsoft azure", "azure devops", "azure functions"],
    "docker": ["docker", "dockerfile", "containerization", "docker compose"],
    "kubernetes": ["kubernetes", "k8s", "helm chart", "container orchestration"],
    "ci/cd": ["ci/cd", "cicd", "continuous integration", "continuous deployment", "continuous delivery", "github actions", "gitlab ci", "jenkins", "circle ci", "travis ci"],
    "terraform": ["terraform", "infrastructure as code", "iac"],
    "ansible": ["ansible"],
    "linux": ["linux", "ubuntu", "centos", "debian", "unix", "linux cli", "terminal linux", "bash scripting"],
    "nginx": ["nginx"],
    "apache": ["apache server", "apache http"],
    "devops": ["devops", "dev ops", "sre", "site reliability"],

    # ==========================================================================
    # UI/UX DESIGN
    # ==========================================================================
    "figma": ["figma"],
    "adobe xd": ["adobe xd"],
    "sketch": ["sketch app", "sketch"],
    "invision": ["invision", "in vision"],
    "prototyping": ["prototyping", "prototype", "wireframe", "wireframing", "mockup"],
    "ux research": ["ux research", "user research", "usability testing", "user testing", "user interview", "heuristic evaluation", "user persona"],
    "ux design": ["ux design", "user experience design", "interaction design", "user experience", "ui ux", "ui/ux design", "uiux", "ui dan ux", "ux designer", "ui designer"],
    "design thinking": ["design thinking"],
    "user journey": ["user journey", "customer journey map", "empathy map", "affinity diagram"],
    "information architecture": ["information architecture", "site map", "navigation design"],
    "interaction design": ["interaction design", "microinteraction", "motion design"],
    "design system": ["design system", "component library", "style guide", "atomic design"],

    # ==========================================================================
    # GRAPHIC DESIGN & CREATIVE TOOLS
    # ==========================================================================
    "adobe photoshop": ["adobe photoshop", "photoshop"],
    "adobe illustrator": ["adobe illustrator", "illustrator"],
    "adobe indesign": ["adobe indesign", "indesign"],
    "adobe premiere pro": ["adobe premiere", "premiere pro"],
    "adobe after effects": ["adobe after effects", "after effects"],
    "coreldraw": ["coreldraw", "corel draw"],
    "canva": ["canva"],
    "blender": ["blender"],
    "cinema 4d": ["cinema 4d", "c4d"],
    "davinci resolve": ["davinci resolve", "da vinci resolve"],
    "lightroom": ["lightroom", "adobe lightroom"],
    "sketchup": ["sketchup", "sketch up"],
    "autocad": ["autocad", "auto cad"],
    "graphic design": ["graphic design", "desain grafis", "design grafis", "visual design", "tipografi", "typography", "design software"],
    "illustration": ["illustration", "ilustrasi", "key visual", "visual communication", "komunikasi visual", "digital illustration", "hand lettering"],
    "logo design": ["logo design", "brand identity design", "visual identity", "branding design"],
    "video editing": ["video editing", "video editor", "pengeditan video", "color grading", "color correction"],
    "photography": ["photography", "fotografi", "photo editing"],
    "3d modeling": ["3d modeling", "3d modelling", "3d design", "3d render", "rendering"],

    # ==========================================================================
    # DIGITAL MARKETING
    # ==========================================================================
    "digital marketing": ["digital marketing", "pemasaran digital", "online marketing", "internet marketing", "marketing online"],
    "seo": ["seo", "search engine optimization", "on-page seo", "off-page seo", "technical seo", "keyword research"],
    "sem": ["sem", "search engine marketing", "pay per click", "ppc"],
    "google ads": ["google ads", "google adwords", "google ad", "adwords"],
    "meta ads": ["facebook ads", "instagram ads", "meta ads", "fb ads"],
    "tiktok ads": ["tiktok ads", "tiktok for business", "tiktok advertising"],
    "email marketing": ["email marketing", "email campaign", "mailchimp", "newsletter"],
    "content marketing": ["content marketing", "strategi konten", "content creation", "content planning"],
    "content strategy": ["content strategy", "editorial strategy", "content roadmap"],
    "social media marketing": ["social media marketing", "manajemen sosial media", "sosial media marketing", "smm"],
    "social media management": ["social media management", "pengelolaan sosial media"],
    "marketing communications": ["marketing communications", "marcomm", "marketing comm"],
    "performance marketing": ["performance marketing", "conversion rate", "cpa", "cpl", "roas"],
    "affiliate marketing": ["affiliate marketing", "affiliate"],
    "influencer marketing": ["influencer marketing", "kol", "key opinion leader"],
    "brand management": ["brand management", "brand strategy", "brand identity", "brand awareness", "brand equity"],
    "market research": ["market research", "riset pasar", "competitive analysis", "competitor analysis", "consumer insight", "market analysis", "analisis pasar", "tren pasar", "category marketing", "customer insights"],
    "growth hacking": ["growth hacking", "growth marketing", "growth strategy", "user acquisition"],
    "trade marketing": ["trade marketing", "modern trade", "traditional trade", "channel marketing", "fmcg", "fast moving consumer goods", "msl", "merchandising program"],
    "brand activation": ["brand activation", "aktivasi merek", "event activation", "field activation", "in-store activation"],
    "visual merchandising": ["visual merchandising", "planogram", "merchandise display", "store layout", "retail display", "posm"],
    "field operations": ["field canvassing", "canvassing", "outlet visit", "kunjungan outlet", "field work", "lapangan", "roadshow", "face-to-face marketing"],

    # ==========================================================================
    # SOCIAL MEDIA
    # ==========================================================================
    "social media": ["social media", "sosial media", "instagram", "linkedin", "facebook", "tiktok", "youtube channel"],
    "content creation": ["content creation", "content creator", "pembuatan konten", "content calendar", "editorial calendar"],
    "community management": ["community management", "community manager", "online community"],

    # ==========================================================================
    # SALES & CRM
    # ==========================================================================
    "salesforce": ["salesforce", "sales cloud", "crm salesforce"],
    "hubspot": ["hubspot"],
    "crm": ["crm", "customer relationship management"],
    "sales": ["sales strategy", "sales target", "sales funnel", "sales pipeline", "lead generation", "closing deal", "b2b sales", "b2c sales", "penjualan produk", "purna jual"],
    "relationship management": ["relationship management", "relationship manager", "account management", "key account management", "wealth management", "private banking", "hnwi", "client portfolio", "portfolio management"],
    "negotiation": ["negotiation", "negosiasi"],
    "partnership": ["partnership development", "partnership management", "partnership marketing", "business partnership", "kerjasama", "mitra", "strategic alliance"],
    "business development": ["business development", "pengembangan bisnis", "pengembangan pasar", "perluasan pasar", "new market", "prospek klien", "akuisisi klien"],

    # ==========================================================================
    # PROJECT MANAGEMENT & AGILE
    # ==========================================================================
    "agile": ["agile", "agile methodology", "agile development"],
    "scrum": ["scrum", "sprint planning", "sprint review", "daily standup", "scrum master"],
    "kanban": ["kanban"],
    "jira": ["jira", "atlassian jira"],
    "trello": ["trello"],
    "notion": ["notion"],
    "project management": ["project management", "manajemen proyek", "project planning", "gantt chart", "project charter"],
    "okr": ["okr", "objective and key result"],
    "pmp": ["pmp certification", "project management professional"],

    # ==========================================================================
    # FINANCE & ACCOUNTING
    # ==========================================================================
    "financial modeling": ["financial modeling", "financial model", "financial analysis", "dcf", "valuation"],
    "accounting": ["accounting", "akuntansi", "general ledger", "accounts payable", "accounts receivable", "journal entry"],
    "sap": ["sap", "sap erp", "sap abap", "sap fico", "sap mm", "sap sd", "sap hana"],
    "erp": ["erp", "enterprise resource planning", "oracle erp", "odoo"],
    "audit": ["auditing", "audit", "internal audit", "financial audit"],
    "taxation": ["pajak", "taxation", "bpjs", "pph", "ppn"],
    "banking operations": ["cash management system", "trade finance", "digital banking", "wholesale banking", "commercial banking", "retail banking", "islamic banking", "bank syariah", "kredit", "lending", "pembiayaan", "perbankan"],
    "actuarial": ["actuarial", "aktuaria", "profit testing", "insurance pricing", "reserving", "life insurance", "general insurance", "asuransi"],

    # ==========================================================================
    # HR & PEOPLE
    # ==========================================================================
    "recruitment": ["recruitment", "recruiting", "talent acquisition", "headhunting", "sourcing kandidat", "job posting", "rekrutmen", "rekrut"],
    "hr management": ["human resource", "hr management", "hrm", "employee relations", "industrial relations"],
    "training & development": ["training and development", "learning and development", "l&d", "employee training", "pelatihan", "training program", "training design", "program pelatihan"],
    "training needs analysis": ["training needs analysis", "tna", "competency management", "people development", "learning management system", "lms", "kirkpatrick", "training effectiveness"],
    "payroll": ["payroll", "penggajian", "kompensasi"],
    "performance management": ["performance management", "performance review", "kpi", "key performance indicator", "key performance indicators"],

    # ==========================================================================
    # CYBERSECURITY
    # ==========================================================================
    "cybersecurity": ["cybersecurity", "cyber security", "information security", "infosec"],
    "penetration testing": ["penetration testing", "pentest", "ethical hacking", "vulnerability assessment"],
    "network security": ["network security", "firewall", "vpn", "ids/ips"],
    "siem": ["siem", "splunk", "security operations", "soc analyst"],
    "cryptography": ["cryptography", "encryption", "ssl/tls", "pki"],

    # ==========================================================================
    # NETWORKING & SYSADMIN
    # ==========================================================================
    "networking": ["networking", "tcp/ip", "dns", "dhcp", "routing", "switching", "cisco", "network infrastructure"],
    "system administration": ["system administrator", "sysadmin", "windows server", "active directory", "ldap"],
    "it support": ["it support", "helpdesk", "technical support", "service desk", "troubleshooting"],

    # ==========================================================================
    # TESTING & QA
    # ==========================================================================
    "software testing": ["software testing", "unit testing", "integration testing", "manual testing", "qa testing", "test case", "bug reporting", "regression testing"],
    "test automation": ["test automation", "automated testing", "selenium", "cypress", "jest", "junit", "pytest", "playwright", "appium"],
    "api testing": ["api testing", "postman", "insomnia", "swagger"],

    # ==========================================================================
    # VERSION CONTROL & TOOLS
    # ==========================================================================
    "git": ["git", "github", "gitlab", "bitbucket", "version control", "git flow", "git repository"],
    "confluence": ["confluence", "atlassian confluence"],

    # ==========================================================================
    # SOFT SKILLS & GENERAL
    # ==========================================================================
    "communication": ["communication skills", "kemampuan komunikasi", "public speaking", "presentasi", "presentation skills", "komunikasi"],
    "teamwork": ["teamwork", "team player", "kolaborasi", "collaboration", "cross-functional"],
    "problem solving": ["problem solving", "analytical thinking", "critical thinking", "pemecahan masalah", "systems thinking"],
    "time management": ["time management", "manajemen waktu", "deadline oriented", "multitasking"],
    "leadership": ["leadership", "kepemimpinan", "team leader", "mentoring", "coaching", "team lead"],
    "adaptability": ["adaptability", "adaptif", "fast learner", "belajar cepat"],
    "attention to detail": ["attention to detail", "teliti", "detail oriented", "ketelitian"],
    "english proficiency": ["english proficiency", "bahasa inggris", "english communication", "toefl", "ielts", "bilingual", "good english", "fluent english", "english writing", "english speaking"],
    "creativity": ["creativity", "creative thinking"],
    "product management": ["product management", "product roadmap", "user story", "backlog", "product lifecycle", "go-to-market", "product owner", "product strategy"],

    # ==========================================================================
    # SOFTWARE DEVELOPMENT GENERAL
    # ==========================================================================
    "software development": ["software development", "web development", "application development", "pengembangan aplikasi", "pengembangan sistem", "coding", "debugging", "software architecture", "scalability", "systems architecture", "memory management"],
    "frontend development": ["front end developer", "frontend developer", "front-end developer", "frontend development"],

    # ==========================================================================
    # DOMAIN-SPECIFIC
    # ==========================================================================
    "blockchain": ["blockchain", "web3", "smart contract", "solidity", "ethereum", "nft", "defi", "cryptocurrency", "crypto"],
    "iot": ["iot", "internet of things", "embedded system", "arduino", "raspberry pi", "mqtt"],
    "ar/vr": ["augmented reality", "virtual reality", "ar/vr", "mixed reality", "unity xr"],
    "game development": ["game development", "unity engine", "unreal engine", "game design", "game mechanic", "level design"],
    "e-commerce": ["e-commerce", "ecommerce", "marketplace", "tokopedia", "shopee", "lazada", "online store"],
    "supply chain": ["supply chain", "rantai pasok", "logistics", "inventory management", "procurement", "warehouse"],
    "ux writing": ["ux writing", "microcopy", "copywriting", "content writing", "technical writing", "dokumentasi teknis"],
    "operations": ["operations management", "process improvement", "lean", "six sigma", "sop", "standard operating procedure", "kpi monitoring", "business efficiency", "risk management"],
    "legal": ["legal research", "legal analysis", "legal compliance", "corporate legal", "hukum", "regulasi", "regulatory compliance", "contract drafting", "legal review", "peraturan", "kepatuhan"],
    "event management": ["event management", "event organizer", "event strategy", "pengelolaan event", "mall promotion", "event planning", "event execution"],
    "proofreading": ["proofreading", "copyediting", "copy editing", "grammar editing", "subtitle", "subtitling", "localization", "translation", "penerjemahan"],
    "creative writing": ["creative writing", "penulisan kreatif", "storytelling", "content evaluation", "writing evaluation", "editorial feedback"],
    "maritime": ["ship building", "galangan kapal", "maritime", "docking", "shipyard", "kapal", "naval", "perkapalan", "teknik kapal"],
    "medical/life science": ["medical device", "life science", "pharmaceutical", "healthcare", "alat kesehatan", "clinical", "medis"],
    "merchandising": ["merchandiser", "merchandising", "display produk", "penataan display", "product display", "sampling produk", "stock monitoring", "stok produk"],
    "stakeholder management": ["stakeholder management", "stakeholder engagement", "internal stakeholders", "cross-team", "cross functional", "koordinasi divisi", "koordinasi antar tim"],
    "vendor management": ["vendor management", "vendor selection", "vendor evaluation", "procurement vendor", "managing vendor"],

    # ==========================================================================
    # SUSTAINABILITY, ENGINEERING & RESEARCH
    # ==========================================================================
    "sustainable development": ["sustainable development", "sustainable design", "sustainable systems", "sustainable technologies", "sustainable architecture"],
    "civil engineering": ["civil engineering"],
    "environmental science": ["environmental science", "environmental engineering", "natural resource management", "climate change mitigation"],
    "building design": ["building design", "architectural design"],
    "community development": ["community development"],
    "land management": ["land management", "land development"],
    "materials science": ["materials science"],
    "chemistry": ["chemistry"],
    "laboratory testing": ["laboratory testing"],
    "research": ["research", "case studies", "policy analysis", "public policy"],
}

def normalize_text(text: str) -> str:
    text = str(text).lower()
    text = text.replace("/", " ")
    text = re.sub(r"[^a-z0-9+#.\-\s]", " ", text)
    text = re.sub(r"\s+", " ", text).strip()
    return text

def extract_skills_from_text(text: str, skill_map: Dict[str, List[str]]) -> Set[str]:
    normalized = normalize_text(text)
    found_skills = set()
    for canonical_skill, aliases in skill_map.items():
        for alias in aliases:
            alias_norm = normalize_text(alias)
            if len(alias_norm) <= 2:
                pattern = r"(?<![a-z0-9+#.])" + re.escape(alias_norm) + r"(?![a-z0-9+#.])"
            else:
                pattern = r"\b" + re.escape(alias_norm) + r"\b"
            if re.search(pattern, normalized):
                found_skills.add(canonical_skill)
                break
    return found_skills

def detect_focus_area(skill: str) -> str:
    ai_data = {
        "machine learning", "deep learning", "tensorflow", "pytorch", "scikit-learn",
        "pandas", "numpy", "data analysis", "data visualization", "nlp", "computer vision",
        "statistics", "big data", "mlops", "etl", "data warehouse", "power bi", "tableau",
        "looker", "google analytics", "business intelligence"
    }
    software = {
        "python", "java", "javascript", "typescript", "php", "go", "c++", "c#", "c", "rust",
        "ruby", "swift", "dart", "scala", "r", "cobol", "bash/shell", "matlab", "groovy",
        "kotlin", "html", "css", "react.js", "vue.js", "angular", "svelte", "jquery",
        "webpack", "vite", "node.js", "laravel", "django", "flask", "fastapi", "spring boot",
        "asp.net", "ruby on rails", "codeigniter", "symfony", "rest api", "graphql", "grpc",
        "websocket", "microservices", "wordpress", "magento", "shopify", "sql", "nosql",
        "mysql", "postgresql", "mongodb", "redis", "sqlite", "oracle db", "microsoft sql server",
        "elasticsearch", "cassandra", "dynamodb", "firebase", "git", "software development",
        "frontend development", "android", "ios", "flutter", "react native", "jetpack",
        "software testing", "test automation", "api testing"
    }
    devops = {
        "docker", "kubernetes", "aws", "google cloud", "microsoft azure", "ci/cd",
        "terraform", "ansible", "linux", "nginx", "apache", "devops", "networking",
        "system administration", "it support", "cybersecurity", "penetration testing",
        "network security", "siem", "cryptography"
    }
    design = {
        "figma", "adobe xd", "sketch", "invision", "prototyping", "ux research", "ux design",
        "design thinking", "user journey", "information architecture", "interaction design",
        "design system", "adobe photoshop", "adobe illustrator", "adobe indesign",
        "adobe premiere pro", "adobe after effects", "coreldraw", "canva", "blender",
        "cinema 4d", "davinci resolve", "lightroom", "sketchup", "autocad", "graphic design",
        "illustration", "logo design", "video editing", "photography", "3d modeling"
    }
    marketing = {
        "digital marketing", "seo", "sem", "google ads", "meta ads", "tiktok ads",
        "email marketing", "content marketing", "content strategy", "social media marketing",
        "social media management", "marketing communications", "performance marketing",
        "affiliate marketing", "influencer marketing", "brand management", "market research",
        "growth hacking", "trade marketing", "brand activation", "visual merchandising",
        "field operations", "social media", "content creation", "community management"
    }
    business = {
        "project management", "communication", "leadership", "problem solving", "product management",
        "agile", "scrum", "kanban", "jira", "trello", "notion", "okr", "pmp", "teamwork",
        "time management", "adaptability", "attention to detail", "english proficiency",
        "creativity", "stakeholder management", "vendor management", "event management",
        "operations", "legal", "proofreading", "creative writing", "ux writing",
        "crm", "salesforce", "hubspot", "sales", "relationship management", "negotiation",
        "partnership", "business development",
        "financial modeling", "accounting", "sap", "erp", "audit", "taxation",
        "banking operations", "actuarial", "excel",
        "recruitment", "hr management", "training & development", "training needs analysis",
        "payroll", "performance management"
    }
    domain = {
        "blockchain", "iot", "ar/vr", "game development", "e-commerce", "supply chain",
        "maritime", "medical/life science", "merchandising", "sustainable development",
        "civil engineering", "environmental science", "building design", "community development",
        "land management", "materials science", "chemistry", "laboratory testing", "research"
    }

    if skill in ai_data: return "Data/AI"
    if skill in software: return "Software Engineering"
    if skill in devops: return "Cloud/DevOps"
    if skill in design: return "Design & Creative"
    if skill in marketing: return "Marketing"
    if skill in business: return "Business & Soft Skill"
    if skill in domain: return "Domain Specific"
    return "General"

def build_rule_based_skill_gap_recommendation(

    model_probability: float,

    predicted_text: str,

    cv_text: str,

    job_text: str,

    skill_map: Dict[str, List[str]] = CANONICAL_SKILL_MAP

) -> Dict[str, Any]:
    cv_skills = extract_skills_from_text(cv_text, skill_map)
    job_skills = extract_skills_from_text(job_text, skill_map)

    matched_skills = sorted(cv_skills.intersection(job_skills))
    missing_skills = sorted(job_skills.difference(cv_skills))
    extra_skills = sorted(cv_skills.difference(job_skills))

    if len(job_skills) > 0:
        rule_skill_coverage = len(matched_skills) / len(job_skills)
    else:
        rule_skill_coverage = 0.0

    hybrid_readiness_score = (0.50 * model_probability) + (0.50 * rule_skill_coverage)

    if hybrid_readiness_score >= 0.80:
        readiness_level = "Sangat Kuat"
    elif hybrid_readiness_score >= 0.65:
        readiness_level = "Cukup Kuat"
    elif hybrid_readiness_score >= 0.50:
        readiness_level = "Perlu Penguatan"
    else:
        readiness_level = "Belum Siap"

    grouped_gaps = {}
    for skill in missing_skills:
        area = detect_focus_area(skill)
        grouped_gaps.setdefault(area, []).append(skill)

    if not missing_skills:
        main_recommendation = (
            "Kandidat sudah memiliki skill utama yang relevan dengan kebutuhan lowongan. "
            "Fokus berikutnya adalah memperkuat portofolio, bukti proyek, dan pengalaman praktis."
        )
        learning_plan = [
            "Minggu 1-2: Rapikan CV dengan bukti proyek, metrik capaian, dan link portofolio.",
            "Minggu 3-4: Latihan interview teknis dan studi kasus sesuai role yang dilamar."
        ]
        cv_improvements = [
            "Pertahankan kata kunci skill yang sudah relevan dan tambahkan dampak terukur dari setiap pengalaman."
        ]
    else:
        priority_skills = missing_skills[:5]
        main_recommendation = (
            "Kandidat perlu memprioritaskan penguatan skill berikut: "
            + ", ".join(priority_skills)
            + ". Skill tersebut muncul pada kebutuhan pekerjaan tetapi belum terdeteksi kuat pada profil CV."
        )
        learning_plan = []
        for week, skill in enumerate(priority_skills[:4], start=1):
            learning_plan.append(f"Minggu {week}: Pelajari dan buat mini project terkait '{skill}'.")
        cv_improvements = [
            "Tambahkan pengalaman, sertifikasi, atau project yang membuktikan skill gap prioritas.",
            "Gunakan kata kunci skill yang relevan dengan lowongan agar ATS/recruiter lebih mudah membaca kecocokan."
        ]

    return {
        "model_probability": model_probability,
        "predicted_text": predicted_text,
        "cv_skills": sorted(cv_skills),
        "job_skills": sorted(job_skills),
        "matched_skills": matched_skills,
        "missing_skills": missing_skills,
        "extra_skills": extra_skills,
        "rule_skill_coverage": rule_skill_coverage,
        "hybrid_readiness_score": hybrid_readiness_score,
        "readiness_level": readiness_level,
        "grouped_gaps": grouped_gaps,
        "main_recommendation": main_recommendation,
        "learning_plan": learning_plan,
        "cv_improvements": cv_improvements
    }


# === HACK KERAS: MENGABAIKAN QUANTIZATION_CONFIG ===
# Model ini dilatih di Colab dengan Keras versi sangat baru, 
# sementara server HF menggunakan versi lebih lama.
# Kode ini memaksa Keras untuk mengabaikan parameter yang tidak dia kenal.
original_layer_init = tf.keras.layers.Layer.__init__

def patched_layer_init(self, *args, **kwargs):
    kwargs.pop("quantization_config", None)
    original_layer_init(self, *args, **kwargs)

tf.keras.layers.Layer.__init__ = patched_layer_init
# ===================================================

# === DEFINISI CUSTOM LAYER ===
# Wajib ditaruh di sini agar file .keras bisa mengenali layer buatan temanmu
@tf.keras.utils.register_keras_serializable(package="TalentIQ")
class TalentIQFeatureInteraction(tf.keras.layers.Layer):
    def __init__(self, include_cosine: bool = True, **kwargs):
        super().__init__(**kwargs)
        self.include_cosine = include_cosine

    def call(self, inputs, training=None):
        cv_vector, job_vector = inputs
        abs_diff = tf.abs(cv_vector - job_vector)
        squared_diff = tf.square(cv_vector - job_vector)
        product = cv_vector * job_vector
        features = [abs_diff, squared_diff, product]
        if self.include_cosine:
            cv_norm = tf.nn.l2_normalize(cv_vector, axis=-1)
            job_norm = tf.nn.l2_normalize(job_vector, axis=-1)
            cosine = tf.reduce_sum(cv_norm * job_norm, axis=-1, keepdims=True)
            features.append(cosine)
        return tf.concat(features, axis=-1)

    def get_config(self):
        config = super().get_config()
        config.update({"include_cosine": self.include_cosine})
        return config

@tf.keras.utils.register_keras_serializable(package="TalentIQ")
class TalentIQGatedDense(tf.keras.layers.Layer):
    def __init__(self, units: int, activation: str = "relu", dropout_rate: float = 0.0, **kwargs):
        super().__init__(**kwargs)
        self.units = units
        self.activation = tf.keras.activations.get(activation)
        self.dropout_rate = dropout_rate
        self.projection = tf.keras.layers.Dense(units, activation=self.activation)
        self.gate = tf.keras.layers.Dense(units, activation="sigmoid")
        self.dropout = tf.keras.layers.Dropout(dropout_rate)
        self.norm = tf.keras.layers.LayerNormalization(epsilon=1e-6)

    def call(self, inputs, training=None):
        projected = self.projection(inputs)
        gate_values = self.gate(inputs)
        gated_output = projected * gate_values
        gated_output = self.dropout(gated_output, training=training)
        return self.norm(gated_output)

    def get_config(self):
        config = super().get_config()
        config.update({
            "units": self.units,
            "activation": tf.keras.activations.serialize(self.activation),
            "dropout_rate": self.dropout_rate
        })
        return config
# =============================

# Inisialisasi API
app = FastAPI(title="TalentIQ AI Matcher")

# Lokasi file Keras (Kita buat dinamis, bisa baca dari folder Model atau root)
if os.path.exists("Model/talentiq_model.keras"):
    MODEL_PATH = "Model/talentiq_model.keras"
elif os.path.exists("talentiq_model.keras"):
    MODEL_PATH = "talentiq_model.keras"
else:
    MODEL_PATH = "model.keras"

model = None
encoder = None
model_load_error = None

print(f"Mencoba memuat model dari: {MODEL_PATH}")
try:
    # Memuat model Keras melalui TensorFlow
    model = tf.keras.models.load_model(MODEL_PATH)
    # Memuat model NLP untuk mengubah Teks -> Angka
    encoder = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')
    print("✅ Model AI & NLP berhasil dimuat!")
except Exception as e:
    import traceback
    model_load_error = traceback.format_exc()
    print(f"⚠️ Error memuat model:\n{model_load_error}")


# Format Input dari Node.js (Sekarang menerima TEKS, bukan list angka)
class PredictRequest(BaseModel):
    cv_text: str
    job_text: str
    threshold: float = 0.5


@app.post("/predict")
def predict_match(req: PredictRequest):
    if model is None or encoder is None:
        raise HTTPException(status_code=500, detail=f"Model gagal dimuat saat server menyala. Error detail:\n{model_load_error}")

    try:
        # 1. Ubah teks dari Node.js menjadi list 384 angka (Embedding)
        cv_emb = encoder.encode(req.cv_text)
        job_emb = encoder.encode(req.job_text)

        # 2. WAJIB DI-NORMALIZE (L2) karena model dilatih dengan data yang sudah di-normalize!
        # Jika tidak, nilai probabilitas akan selalu 0.999 karena nilainya kebesaran.
        cv_norm = cv_emb / np.linalg.norm(cv_emb)
        job_norm = job_emb / np.linalg.norm(job_emb)

        # 3. Sesuaikan bentuk array agar diterima oleh model Keras: format (1, 384)
        cv_input = np.expand_dims(cv_norm, axis=0)
        job_input = np.expand_dims(job_norm, axis=0)

        # 4. Lakukan prediksi kecocokan
        prediction = model.predict([cv_input, job_input])
        
        # Ambil skor probabilitas (contoh: 0.85)
        prob = float(prediction[0][0])
        
        # Tentukan hasil akhir (Match atau Tidak) berdasarkan model DL saja sebagai label awal
        is_match = prob >= req.threshold
        text_label = "Match" if is_match else "Not Match"

        # Analisis Skill Gap
        gap_report = build_rule_based_skill_gap_recommendation(
            model_probability=prob,
            predicted_text=text_label,
            cv_text=req.cv_text,
            job_text=req.job_text
        )

        return gap_report

    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))


@app.get("/")
def health_check():
    return {"message": "TalentIQ AI API (dengan text to embedding) siap digunakan!"}