Abstract

Introduction

Imagine a recruiter overwhelmed with hundreds of resumes for a single job posting. Hours are spent scanning CVs, checking skills, and shortlisting candidates—only to risk missing out on the perfect hire.

Now, picture a team of AI agents working together behind the scenes: one agent cleans and parses resumes, another matches them against job descriptions, while a third highlights strengths and gaps. Finally, all the insights flow into a simple dashboard where recruiters see ranked candidates at a glance.

This is what the Multi-Agent AI Recruitment System does—turning a messy hiring process into a smooth, intelligent, and automated workflow.

✅Here are four points on how this recruitment agent is useful:

1. Speeds Hiring-Automates resume parsing and matching to cut screening time from hours to minutes.
2. Improves Match Quality -Uses AI embeddings and scoring to align candidates with job requirements accurately.
3. Reduces Bias-Redacts personal information to ensure fairer candidate evaluation.
4. Provides Clear Insights-Highlights strengths, gaps, and recommendations for smarter decisions.

Methodology

The Multi-Agent AI Recruitment System works as a pipeline of specialized agents. Each agent performs a specific role using the best-suited tools and models. The output of one agent becomes the input for the next, ensuring a smooth and automated process.

Figure 1: Orchestration of the agents

Agents

1. Resume parsing

As shown in Figure 1 above, the first step is resume parsing. This agent cleans the resume by removing personal information using custom PII detection and redaction tools.

Tools and Models:

• Python Regex (re) for pattern matching.
• Named Entity Recognition model: dslim/bert-base-NER via HuggingFace Transformers.
• Custom Personally Identifiable Information (PII) detection and redaction tools (scan_pii, redact_pii).

Output

• A structured dictionary containing cleaned resume text, extracted skills, experience years, education, and processing messages.

{
  "resume_text": "<cleaned resume text>",
  "extracted_skills": ["Python", "Data Analysis", "SQL"],
  "experience_years": 5,
  "education": "Bachelor",
  "messages": ["PII detected and redacted", "Parsed: 5 skills, 5 yrs exp"]
}

2. Embedding and matching

Next is the embedding and matching step. Cosine similarity is computed using PyTorch to produce a match score expressed as a percentage. This score reflects how well the candidate fits the job requirements.

Tools and Models:

• sentence-transformers/all-MiniLM-L6-v2 for embedding generation.
• PyTorch (torch.cosine_similarity) for similarity computation.

Output

• A dictionary containing the match score and processing messages.

{
  "match_score": 85.4,
  "messages": ["Match score: 85.4%"]
}

3. Insight generation

The third step is insight generation. This agent decides whether a candidate should be shortlisted by checking if the match score exceeds a threshold (70%).

Tools and Models:

• Python string and list operations for logic and keyword matching.

Output

• A dictionary with strengths, missing skills, shortlist status, and processing messages.

{
  "strengths": ["Python", "Data Analysis", "SQL"],
  "missing_skills": ["Banking experience"],
  "shortlisted": true,
  "messages": ["Shortlisted: True"]
}

4. Orchestration connects all agents into a sequence

Finally, the workflow orchestration connects all agents into a sequence. This is done using LangGraph’s StateGraph, which manages state and message flow between agents. The process starts at the resume parser, moves to the embedder, and finishes with the insight generator. The final output is a comprehensive candidate evaluation that includes all extracted and computed information, along with a set of messages documenting the process.

Tools and Models:

• StateGraph from LangGraph for workflow orchestration.
• Python TypedDict for structured state representation (RecruitmentState)

Final Output:

• A comprehensive state dictionary containing all processed candidate information and evaluation results.

{
  "candidate_name": "John Doe",
  "candidate_email": "johndoe@example.com",
  "resume_text": "<cleaned resume text>",
  "job_description": "...",
  "job_requirements": "...",
  "extracted_skills": ["Python", "Data Analysis"],
  "experience_years": 5,
  "education": "Bachelor",
  "match_score": 85.4,
  "strengths": ["Python", "Data Analysis"],
  "missing_skills": ["Banking experience"],
  "shortlisted": true,
  "messages": ["PII detected and redacted", "Parsed: 5 skills, 5 yrs exp", "Match score: 85.4%", "Shortlisted: True"]
}

Installation & Setup

Prerequisites:

• Python 3.10+
• Node.js 18+ (for frontend React)
• npm or yarn (frontend package manager)
• PostgreSQL (for production) or SQLite (for demo/testing)
• Git (for cloning the repository)

Clone the Repository:

git clone https://github.com/your-username/hr-recruitment-agent.git
cd hr-recruitment-agent

Backend Setup:

1. Create and activate a Python virtual environment:
   • Linux/macOS: python -m venv venv && source venv/bin/activate
   • Windows: python -m venv venv && venv\Scripts\activate
2. Install required packages: pip install -r requirements.txt
3. Configure environment variables:
   • SECRET_KEY, DEBUG, DATABASE_URL, HF_SPACE_API_URL
4. Run migrations: python backend/manage.py migrate
5. Create an admin user: python backend/manage.py createsuperuser
6. Collect static files: python backend/manage.py collectstatic --noinput
7. Run development server: python backend/manage.py runserver 8084

Frontend Setup:

1. Navigate to frontend directory: cd frontend
2. Install dependencies: npm install
3. Run frontend server: npm start
4. Access application: Frontend: http://localhost:3000, Admin: http://localhost:8084/admin, API: http://localhost:8084/api/jobs/

Options:

• Render Full-Stack Deployment (Recommended)
• Hugging Face Spaces (Demo)

User Instructions

Job Management: create/update jobs, assign departments and requirements.
Resume Upload & Candidate Screening: upload resumes, AI extracts skills, experience, performs semantic matching, and redacts PII.
Candidate Scoring & Shortlisting: match percentage, auto-shortlisting, strengths/gap analysis, dashboard visualization.
Interview Scheduling: automatic email notifications and HR adjustments.
Human-in-the-Loop Review: human validation and feedback integration.

Maintenance & Support
Updating dependencies, database management, AI model maintenance, monitoring and logs.

Error Handling & Resilience
Retry failed tasks, fallback mechanisms, error notifications, health checks.

Performance Evaluation
Track AI accuracy, processing times, benchmarking, and analytics logging.

Security & Data Privacy
HTTPS, encrypted storage, PII testing, audit logs, GDPR compliance.

Testing & Quality Assurance
Unit, integration, end-to-end tests, CI/CD pipelines.

User Experience & Accessibility
Responsive UI, accessibility support, onboarding tutorials, feedback collection.

Troubleshooting
Backend startup issues, database connection errors, resume scoring failures, email issues, dashboard update problems.

Best Practices
Regular updates, backups, monitoring, HR staff training, clear workflow documentation.