1. Project Overview

The Enterprise Intelligent Recruitment System is a complete enterprise-level AI intelligent recruitment management platform.

It adopts a microservice architecture design, deeply integrating traditional recruitment business (job posting, resume screening, interview, offer, contract, onboarding) with artificial intelligence, covering the entire recruitment lifecycle from job creation and candidate outreach to employee onboarding.

The system has built-in capabilities such as AgentScope multi-agent orchestration, LLM multi-model routing, AI resume parsing and screening, AI intelligent question generation and interview evaluation, offer/retention prediction, AI talent recommendation, data analysis insights, and provides enterprise-level features like contract electronic signatures, message notifications, audit logs, and page watermarks.

More project practices at Java突击队网: susan.net.cn/project

1.1 Project Positioning

Dimension	Description
Project Type	Enterprise-level AI Intelligent Recruitment Management Platform
Architecture Pattern	Microservice Architecture (Spring Cloud + Nacos)
Backend Technology	Spring Boot 4.1 + Java 25 Virtual Threads + Maven
Frontend Technology	Vue 3 + TypeScript + Vite + Element Plus
Database	MySQL 8.0 (7 business databases) + Redis
Service Registry	Nacos Registry + Configuration Center
File Storage	RustFS (S3-compatible object storage)
Search Engine	Elasticsearch (talent pool semantic search, optional)
AI Capabilities	LLM Gateway + AgentScope Multi-Agent Orchestration + Visual Model Resume Parsing
Large Models	Tongyi Qianwen (qwen3-max / qwen-vl-max) + DeepSeek

1.2 Project Scale

Metric	Value
Backend Microservices	7 business services + API Gateway + Common Module
Business Databases	7 (55 business tables, database-per-service)
REST APIs	315+ (Knife4j online debugging)
AI Agents	11 (covering 12 types of AI capabilities)
Multi-Agent Orchestration	4 sub-agent collaborations + Resume→Offer five-stage pipeline
Frontend Applications	2 (Recruitment Management Console + Recruitment Website)
Follow-along Tutorials	30+ feature development documents

1.3 System Architecture Diagram

2. Backend Microservice Architecture

The project adopts a Spring Cloud microservice architecture, comprising a total of 8 Maven modules, implementing service registration and configuration management via Nacos, and a unified API gateway via Spring Cloud Gateway.

2.1 Module List

Module	Service Name	Port	Core Responsibilities
smart-recruit-common	Common Module	-	Common utilities, DTO/VO, exception handling, unified configuration
smart-recruit-gateway	API Gateway	8080	Unified entry, routing, JWT authentication, rate limiting, CORS
smart-recruit-system	System Service	8081	Users/Roles/Permissions/Departments, authentication, system config, notifications, audit logs
smart-recruit-recruitment	Recruitment Service	8082	Job management, resume parsing & AI screening, candidate center, website jobs, dashboard dynamics/todos
smart-recruit-interview	Interview Service	8083	Interview management, AI question generation, question bank, online assessment, AI evaluation report
smart-recruit-offer	Offer Service	8084	Offer approval flow, AI acceptance prediction, contract management (e-signature), onboarding, AI retention prediction
smart-recruit-talent	Talent Service	8085	Talent pool, AI talent recommendation, data analysis (scheduled snapshots + AI insights), dashboard aggregation
smart-recruit-referral	Referral Service	8086	Referral plans/jobs/records, referral leaderboard, candidate submissions
smart-recruit-ai-engine	AI Engine	8087	LLM Gateway (multi-model routing), AgentScope multi-agent, visual resume parsing, Agent task/event/metric monitoring

2.2 Technology Stack

Technology Stack	Version	Purpose
Spring Boot	4.1.0	Application Framework
Spring Cloud	2025.1.2	Microservice Governance
Spring Cloud Alibaba	2025.1.0.0	Nacos Integration
Java	25	Programming Language (Virtual Threads)
Maven	3.9+	Project Build
Spring Cloud Gateway	5.0.2	API Gateway (WebFlux)
MyBatis-Plus	3.5.16	ORM Framework
MySQL	8.0+	Business Database
Redis	6.x/7.x	Cache, Verification Codes, Rate Limiting
Nacos	2.x	Registry + Configuration Center
LangChain4j	0.29.1	LLM Integration Framework
AgentScope	2.0.x	Multi-Agent Orchestration (Interview Evaluation/Orchestration)
Knife4j	4.3.0	API Documentation
JWT (jjwt)	0.12.6	Identity Authentication
RustFS	S3 Compatible	Distributed File Storage
Elasticsearch	8.x	Talent Pool Vector/Semantic Search (Optional)

2.3 Inter-Service Call Relationships

Inter-service communication uses Spring's native @HttpExchange declarative HTTP interface client, combined with @LoadBalanced RestClient and Nacos for service discovery and load balancing, making the interface itself the documentation with zero hand-written HTTP code.

3. Core Functional Modules

3.1 Dashboard

A unified dashboard after login, aggregating full-process data:

• Daily Data: New candidates today/yesterday/this month, number of open positions;
• Process Overview: Pending interviews, pending offers, onboarded this month, average days to start;
• Recruitment funnel and department recruitment progress charts;
• Recent activity (activity stream) and to-do items (default shows latest 20 each, view all available).

3.2 Job Management

Full lifecycle job management:

Function	Description
Job CRUD	Create/Edit/View/Pause/Close positions
Draft & Publish	New jobs default to draft, enter recruitment flow upon publishing, records publisher
AI JD Generation	Asynchronously calls LLM to generate responsibilities/requirements/pluses, timeout does not block page
Job Statistics	Count of jobs by status and active recruitment statistics

3.3 Resume Screening

Resume processing is one of the system's core pipelines:

• Resume parsing supports Word/PDF/Images (images and scans use visual LLM), parsing failures are automatically flagged for manual review;
• AI Screening Switch: Optional during upload, when enabled asynchronously performs intelligent screening, page shows real-time "Screening→Complete" progress and match score;
• Parsing results include structured fields like skills, work experience, project experience, education background, and personal strengths.

3.4 Candidate Center

• Candidate Profile: Basic info, contact details, skill tags, job-seeking status, source channel;
• Full-chain association with resumes, job applications, interviews, and offers;
• Candidate stage history tracking.

3.5 Interview Management

Function	Description
Interview Schedule	Day/Week view schedule, arrange/cancel interviews, supports AI question generation status column
AI Intelligent Question Generation	Asynchronously generates questions by job direction (Frontend/Backend/QA/Algorithm etc.), supports technical/project/behavioral types and question type allocation
Interview Question Bank	Set management (linked to department + position), questions include type/options/answers/interpretations, supports direct population from AI question generation
Online Assessment	Create new assessment (AI generated or selected from bank), send to candidate email, auto-scoring
AI Evaluation Report	AgentScope multi-agent six-dimension evaluation (Skills/Communication/Logic/Collaboration/Learning/Stress Tolerance), both synchronous and streaming modes
Interview Feedback	Interviewer feedback entry and summary

3.6 Offer Management

• Offer creation, multi-level approval flow (approval nodes configured by department);
• AI Acceptance Prediction: Scheduled task calls LLM, predicts acceptance probability from 5 dimensions (salary competitiveness, job match, market conditions, candidate activity, company brand) and gives suggestions, already predicted offers are not called again;
• Offer number, candidate, salary structure management.

3.7 Contract Management

• One-click generation of formal employment contract from Offer (auto-fills company info, contract term, salary, level);
• Electronic Signature: Candidate handwritten signature pad + candidate/HR dual signing process, contract takes effect after signing;
• Contract voiding (fill in void reason), printing, signing records;
• Contract status color coding (Draft/Pending Approval/Sent/Signed/Active/Voided).

3.8 Onboarding Management

• Six-step onboarding process: Document Collection → Equipment Issuance → Account Setup → Welcome Page → Mentor Assignment → Onboarding Training;
• Employee Status: Pending Onboarding/Probation/Regular/Resigned;
• Document verification, equipment issuance records;
• AI Retention Prediction: Scheduled task predicts 6/12 month retention risk based on candidate profile + onboarding process completion + historical job-hopping, skips duplicate calls if process data fingerprint (hash) is unchanged.

3.9 Talent Pool

• Talent pool management (candidate-job matching and entry);
• AI Talent Recommendation: Asynchronously matches and recommends after selecting a job (8 candidates per batch, supports "next batch" pagination), recommendation results cached with Redis 15min + Caffeine 5min dual-level cache;
• Elasticsearch semantic talent search (optional).

3.10 Data Analysis

• Scheduled Task Statistics: Candidate/Interview/Offer daily snapshots pre-aggregated by each module's scheduled tasks, avoiding large real-time queries;
• Data Dashboard: Candidate funnel, department progress, trend analysis;
• AI Intelligent Insights: Calls LLM on statistical results to generate natural language insights.

3.11 Referral Management

• Referral plans, plan positions, share links;
• Referral records, bonus records;
• Referral Leaderboard: Ranked by referral count/bonus;
• Candidates apply via share page.

3.12 Recruitment Website

• Website configuration (Hero area, statistics, culture & values, call-to-action, footer, navigation);
• Hot jobs / Social Recruitment / Campus Recruitment job maintenance (tags, responsibilities, requirements, pluses);
• Independent website frontend (smart-recruit-careers :3001).

3.13 System Management

Function	Description
Users/Roles/Permissions/Departments	RBAC permission model, menu-button-API three-tier permissions
Audit Log	Records who did what and when (operator, IP, parameters, duration)
System Settings	Public config, website config, AI weight config, company info
Page Watermark	Controlled by system setting toggle, displays current login account, name, and time
Message Notifications	In-app messages (auto-notification for business events), unread badge, message detail and jump

3.14 AI Engine (Agent Monitoring)

4. AI Capability Matrix

AI Capability	Business Entry Point	Implementation	Status
Resume Parsing (Text)	Resume Screening → Upload Resume	Tika + LLM Structuring	✅
Resume Parsing (Image/Scan)	Resume Screening → Upload Image Resume	qwen-vl-max Visual Model	✅
AI Intelligent Screening	Resume Screening → Enable AI Screening	LLM Five-Dimension Weighted Scoring + Heuristic Fallback	✅
AI JD Generation	Job Management → New/Edit → AI Generate JD	LLM Async Generation	✅
AI Intelligent Question Generation	AI Question Page / New Set → AI Intelligent Generation	LLM Async Task + Polling	✅
AI Interview Evaluation	Interview Management → AI Evaluation Report	AgentScope HarnessAgent Six Dimensions	✅
Offer Acceptance Prediction	Offer Detail → AI Acceptance Probability Card	Scheduled Task + LLM Five-Dimension Prediction	✅
New Hire Retention Prediction	Onboarding Detail → Retention Prediction	Scheduled Task + Fingerprint Dedup + LLM	✅
AI Talent Recommendation	Talent Pool → Select Job Match	LLM Async Recommendation + Dual-Level Cache	✅
Data Analysis AI Insights	Data Analysis → AI Insights	LLM Generates Natural Language Insights	✅
Multi-Agent Orchestration	AI Engine → Full Process Orchestration	AgentScope Sub-Agent Collaboration	✅
Agent Monitoring	AI Engine → Task Queue/Event Stream/Token	Real Task & Metric Recording + SSE	✅

---

5. Database Design

The system adopts a database-per-service design, with a total of 7 MySQL business databases and 55 business tables.

5.1 Database Distribution

Database	Corresponding Service	Core Tables
smart_recruit_system	System Service	Users, Roles, Permissions, Departments, Notifications, Audit Logs, System Config
smart_recruit_recruitment	Recruitment Service	Jobs, Resumes, Candidates, Applications, Activity Stream, AI Screening Results, Website Jobs
smart_recruit_interview	Interview Service	Interviews, Evaluations, Feedback, Question Bank Sets/Questions, Online Assessments, AI Question Records
smart_recruit_offer	Offer Service	Offers, Approval Flow Config/Records, Contracts, Signing Records, Onboarding/Documents/Equipment
smart_recruit_talent	Talent Service	Talent Pool, Talent Activities, Data Analysis Snapshots
smart_recruit_referral	Referral Service	Referral Plans/Jobs/Records, Bonuses, Leaderboard, Share Tokens
smart_recruit_ai	AI Engine	Agent Info/Metrics/Tasks/Event Logs, AI Parsing Logs

5.2 Design Principles

• Database-per-Service: Each microservice has its own independent database, decoupling at the data layer;
• Snowflake Algorithm Primary Keys: Globally unique Long IDs, supporting distributed scaling;
• Logical Deletion: Core tables uniformly use a deleted field for soft deletion;
• Audit Fields: Six-piece set of create_time / update_time / create_user_id / create_by / update_user_id / update_by;
• Index Specification: Composite indexes cover queries, clean up redundant single-column prefix indexes, avoiding index redundancy;
• JSON Extension: Complex structures (question bank questions, AI results, approval nodes) use JSON columns + TypeHandler.

6. Frontend Architecture

6.1 Technology Stack

Technology	Version	Purpose
Vue	3.5	Core UI Framework
TypeScript	5.5	Type Safety
Vite	6.0	Build Tool
Element Plus	2.9	UI Component Library
Vue Router	4.x	Client-Side Routing
Pinia	2.x	State Management
Axios	1.x	HTTP Client
Chart.js	4.x	Data Visualization
mammoth	-	DOCX Resume/Contract Preview

6.2 Frontend Applications

Application	Port	Description
smart-recruit-frontend	3000	Recruitment Management Console (Dashboard + All Business Modules)
smart-recruit-careers	3001	Recruitment Website (Job Browsing, Application, Contract Signing)

6.3 Main Pages

Category	Pages
Authentication	Login, Register
Dashboard	Dashboard, Recent Activity, To-Do Items
Recruitment	Job Management, Resume Screening, Candidate Center, Interview Management
Interview	AI Intelligent Question Generation, Interview Question Bank, Online Assessment
Offer	Offer Management, Contract Management, Onboarding Management
Talent	Talent Pool, Data Analysis
Referral	Referral Management, Referral Leaderboard
Website	Website Config, Hot Jobs, Social Recruitment, Campus Recruitment
System	User Management, Role Management, Permission Management, Department Management, Audit Log, System Settings
AI Engine	Agent Monitoring
Public	Recruitment Website, Contract Signing Page

6.4 Frontend Engineering Highlights

• Axios Request Interception System: Auto-inject Token, silent 401 refresh + request queue replay, unified business error prompts;
• Route Lazy Loading + Permission Guards: Redirect to login if not authenticated, intercept if unauthorized;
• Element Plus Componentization: Enterprise-grade interactions like tree department selector, paginated tables, progress bars, signature pads;
• Watermark Component: Configurable toggle, globally displays current user and time.

7. System Showcase

More project practices at Java突击队网: susan.net.cn/project

8. Project Highlights

8.1 Architecture Highlights

1. True Microservice Implementation, Not "Just Renaming Modules"

7 business services deployed independently, each with its own independent MySQL business database (7 databases, 55 tables), registered and discovered via Nacos, services calling each other with declarative clients, JWT authentication uniformly converged at the gateway layer—clear service boundaries and call relationships, not simple Maven multi-modules.

① Gateway Unified Authentication, Service Layer Zero Awareness

Gateway AuthFilter completes JWT verification upon request entry, extracts user info and passes it downstream via X-User-Id / X-Username / X-User-Role Headers. Downstream services no longer repeat authentication logic, just read user identity from Headers—separation of concerns, cleaner code.

② Declarative HTTP Client, Interface as Documentation

Inter-service communication uses Spring's native @HttpExchange declarative calls, defining remote calls with interfaces + annotations, completely eliminating HTTP connection code:

@HttpExchange("/api/v1/agent-capabilities")
public interface AiAgentCapabilityClient {
    @PostExchange("/interview/questions")
    ApiResponse<List<Map<String, Object>>> generateInterviewQuestions(...);
}

All services automatically register with Nacos after startup, supporting health checks, load balancing, dynamic up/down; gateway routes all converged into YAML config (spring.cloud.gateway.server.webflux.routes), adding services or paths just requires config changes, zero code.

spring:
  cloud:
    gateway:
      server:
        webflux:
          routes:
            - id: smart-recruit-system
              uri: lb://smart-recruit-system
              predicates:
                - Path=/api/v1/auth/**, /api/v1/users/**, /api/v1/configs/**

2. LLM Gateway Multi-Model Routing + Auto Fallback, Not "Hardcoded to One Model"

All AI calls uniformly go through the AI Engine LLM Gateway, routing models by Agent, Qwen / DeepSeek dual-vendor primary/backup switching, automatic fallback on failure:

Agent	Routed Model
Resume Visual Parsing	qwen-vl-max
AI Screening / AI Question Gen / JD Gen	qwen3-max
Offer Acceptance / Retention Prediction	deepseek-v4-flash
Talent Recommendation / Referral Matching	deepseek-chat

Every call records actual Token consumption, feeding into "Token Consumption Distribution (Last 24h)" monitoring; switching models or adjusting Keys requires no business code changes.

3. AgentScope 2.0 Multi-Agent Orchestration, Not "Single LLM Call"

Interview evaluation is completed by an orchestration master scheduling 4 sub-agents via agent_spawn collaboration, supporting both synchronous and SSE streaming modes:

Sub-agents declared via SubagentDeclaration, master prompt triggers agent_spawn scheduling:

• AgentScope does not directly connect to LLMs; it reuses LangChain4j gateway's unified routing, fallback, and billing through an adapter, completely decoupling the orchestration layer from the model layer;
• Besides sub-agent collaboration, there's also a "Resume → Offer" five-stage full-process pipeline (Parsing → Screening → Manual Review → Interview Evaluation → Offer Prediction), each stage recording real Agent tasks, runnable with one click, fully observable.

4. Visual Model Resume Parsing, Not "Only Text Resumes"

Text resumes go through Tika extraction + LLM structuring;

Image and scanned resumes automatically switch to visual LLM (qwen-vl-max), parsing failures automatically flagged for manual review:

Resume Format	Processing Pipeline	Failure Fallback
PDF / DOCX / DOC Text	Tika Extraction → LLM Structuring	Record parse failure reason
Image / Scanned PDF	qwen-vl-max Visual Parsing	Auto-flag for manual review

Parsing results include structured fields like skills, work experience, project experience, education background, and personal strengths, directly driving subsequent AI screening and talent pool entry.

5. Full-Chain Async + Scheduled Pre-Computation, Not "Page Freezes Waiting for AI"

AI question generation, resume parsing, AI screening, JD generation, talent recommendation are all async task-based: task queue + frontend polling + real-time progress display, no time-consuming call blocks the page; layered with three cost optimizations:

Method	Description
Scheduled Prediction Persistence	Offer acceptance and retention predictions pre-computed by scheduled tasks, detail pages read directly from DB
SHA-256 Fingerprint Dedup	Retention prediction judges by onboarding 5-step data fingerprint, no repeated LLM calls if data unchanged
Dual-Level Cache	Talent recommendation Redis 15min + Caffeine 5min, supports "next batch" pagination

Fingerprint calculation for retention prediction (onboarding 5-step data → SHA-256).

6. Real Agent Monitoring, Not "Hardcoded Demo Data"

Agent task queues, real-time event streams (SSE), Token consumption distribution (last 24h), success rate and health all come from real call records, every Agent task traceable to a specific business operation (who triggered screening on which resume, which interview triggered question generation).

7. Full Electronic Contract Signing Process, Not "Upload an Image"

Candidate handwritten signature pad → email signing link → candidate / HR dual signing → contract effective, signing records, void reasons, print output fully traceable, a truly closed-loop digital contract process.

8. Data Analysis Scheduled Snapshots + AI Insights

Candidate / Interview / Offer daily snapshots pre-aggregated by each module's scheduled tasks, APIs read snapshots directly, completely avoiding real-time statistical timeouts under large data volumes; AI engine then generates natural language insights and optimization suggestions based on statistical results, dashboards and reports have both real data and intelligent interpretation.

9. Java 25 Virtual Threads, High-Concurrency IO Weapon

High IO scenarios like AI async generation, SSE event streams, email sending use virtual threads, significantly reducing resource usage and boosting throughput—one virtual thread per request, scheduled by the OS, hundreds of thousands of concurrent connections are no longer a bottleneck.

10. LangChain4j → AgentScope Model Adaptation, Orchestration and Model Layer Decoupled

AgentScope does not directly connect to LLMs; it reuses LangChain4j gateway's unified routing, fallback, and billing through an adapter, two frameworks each doing their job, seamlessly connected—the orchestration layer only cares about "how multi-agents collaborate", the model layer only cares about "which vendor to call, how to fallback, how many Tokens spent".

11. Multi-Level Fallback Chain, Pages Have Data Despite Any Failure

Interview evaluation, resume screening, Offer/retention prediction all implement multi-level fallback; if LLM fails, Agent fails, or network breaks, pages still have data, exceptions never exposed to users.

Scenario	Fallback Chain
Interview Evaluation	Multi-Agent Orchestration → Single Agent → LLM Direct Call → Heuristic Scoring
Resume Screening	LLM Intelligent Screening → Heuristic Scoring
Offer / Retention Prediction	LLM → Local Agent → Default Value

12. Dual-Level Cache Talent Recommendation, Same Job Returns in Seconds

AI talent recommendation results cached with Redis 15min + Caffeine 5min dual-level cache, supports "next batch" pagination, same job re-query directly hits cache, no repeated Token consumption, no repeated LLM calls.

8.2 Engineering Highlights

Category	Highlight Count	Core Keywords
Microservice Architecture	8	Database-per-service, Gateway Auth Passthrough, Declarative Client, Config-Driven Routing, Virtual Threads, Unified Exception, Common Convergence, One-Click Init
AI / LLM	11	Multi-Model Routing Fallback, AgentScope Multi-Agent, Visual Parsing, Async Tasks, Fingerprint Dedup, Dual Cache, Multi-Level Fallback, Question Bank Linkage, Real Monitoring, AI Insights
Business & Data	7	Resume Pipeline, E-Signature, Six-Step Onboarding Alert, Message Linkage, Scheduled Stats, Security Audit, Data Standards
Frontend Engineering	5	Axios Interception, Route Guards, Dual Frontend, Type Safety, Tutorial System

1. Complete Security and Audit Suite

JWT Authentication + RBAC Three-Tier Permissions (Menu / Button / API) + AOP Operation Audit Log (who, when, from which IP, did what) + Configurable Page Watermark (Account / Name / Time) + Gateway Rate Limiting, complete enterprise compliance capabilities.

2. Data Engineering Standards

Snowflake IDs, Logical Deletion, Audit Six-Piece Set (create/update × time / user ID / username), JSON Extension Columns + TypeHandler, Index Prefix Redundancy Cleanup, table structures standardized and auditable; init.sql one-click initializes 7 databases 55 tables + base data, schema.sql for structure verification anytime.

3. Complete Frontend Infrastructure

Axios request interception (Token injection, silent 401 refresh + request queue replay, unified error prompts), route lazy loading + permission guards, dual frontend architecture (management console :3000 + recruitment website :3001), full TypeScript type safety coverage.

Request 401 ──> Silent Token Refresh ──> Queued Requests Batch Replay ──> User Unaware
Refresh Fails ──> Clear Login State ──> Redirect to Login Page after 1.5s

8.3 What Can This Source Code Bring You?

If you are a junior/mid-level backend developer:

• How to split microservices, how to do service discovery, how to write @HttpExchange declarative interfaces, it's all in this project;
• How to design JWT + RBAC permission models and write the code, look directly at smart-recruit-system;
• MyBatis-Plus advanced usage (pagination, joins, logical deletion, JSON field TypeHandler);
• How to implement Spring Cloud Gateway 5.x routing and authentication (config-driven, configurable whitelist);
• AOP aspects implementing operation audit logs;
• How scheduled tasks + Java 25 virtual thread async are used in real business.

If you are a frontend developer:

• How to organize directories and code for a Vue 3 + TypeScript enterprise project;
• Deep usage of Element Plus (tree select, tables, progress bars, handwritten signature pad);
• Complete solution for Axios interceptors, Token refresh, request queue replay;
• Implementation ideas for public pages like recruitment websites and contract signing;
• How to encapsulate Chart.js data visualization and Word resume preview (mammoth).

If you focus on AI / LLM:

• Complete implementation of LLM gateway multi-model routing + auto fallback;
• AgentScope 2.0 HarnessAgent multi-agent orchestration + agent_spawn collaborative scheduling;
• Implementation of visual LLM parsing image/scanned resumes;
• Engineering practices for async question generation + polling progress, scheduled prediction + fingerprint dedup, dual-level cache recommendation;
• How LangChain4j integrates in a real Spring Boot project;
• A complete set of AI engineering methodologies from "prompt design" to "multi-level fallback".

If you are an architect or tech leader:

• How to split microservices, how to define each service's boundaries, the rationale for 7 databases;
• How to layer AI capabilities (Model Gateway / Orchestration Layer / Business Agent) to be pluggable, fallback-capable, and observable;
• How to model the full recruitment data flow (Resume → Offer → Onboarding);
• What a system where "AI capabilities truly run in every business link" looks like;

More project practices at Java突击队网: susan.net.cn/project

Summary

The Enterprise Intelligent Recruitment System is a feature-complete, modern-architecture, deeply AI-empowered enterprise-level intelligent recruitment management platform: 7 microservices + 7 business databases carry the full recruitment process, 11 AI Agents and AgentScope 2.0 multi-agent orchestration build the AI capability layer, 315+ REST APIs cover 55 business tables, dual frontends open to both HR and candidates.

Core Capability Summary:

• Full Recruitment Closed Loop: Job → Resume → Screening → Interview → Offer → Contract → Onboarding, AI empowerment at every step;
• 11 AI Agents, 12 AI Capabilities: Resume parsing (text + visual), intelligent screening, AI question generation, interview evaluation, JD generation, Offer/retention prediction, talent recommendation, referral matching, data insights, multi-agent orchestration, all async + scheduled pre-computation ensuring performance;
• Multi-Agent Collaboration: AgentScope 2.0 HarnessAgent orchestrates 4 sub-agents to complete interview evaluation, Resume → Offer five-stage pipeline runnable with one click;
• Enterprise Engineering Capabilities: Database-per-service, gateway unified auth, unified response, audit logs, page watermarks, contract e-signatures, complete compliance and security capabilities;
• Visual Operations: Agent task queues, SSE real-time event streams, Token consumption distribution, success rate and health all from real call records;
• Followable, Replicable: 40+ feature development tutorials, step-by-step from zero to full functionality.

This is an intelligent recruitment infrastructure directly usable in enterprise production environments, and also a high-quality reference implementation integrating architecture, AI, and engineering.