🤖 LLM-Based Agentic AI HR Automation System

Project Overview

HR departments spend 60–70% of their time on repetitive, rule-based tasks — screening hundreds of CVs, scheduling interviews, sending follow-up emails. This project replaces that manual workflow with a multi-agent LLM system that automates the entire recruitment pipeline end-to-end.

Problem Statement

A mid-size company receives 200+ applications per job posting. A recruiter manually reads each CV, shortlists candidates, coordinates interview slots across calendars, sends confirmation emails, and follows up. This process takes days and introduces subjective bias.

The goal was to build an agentic system where a user can simply say "Hire a Senior ML Engineer — screen these 150 CVs and schedule interviews for the top 5" and the system handles everything.

Multi-Agent Architecture

Agent Implementation

from langchain.agents import AgentExecutor, create_openai_tools_agent
from langchain_openai import ChatOpenAI
from langchain.tools import tool
import pdfplumber, json

llm = ChatOpenAI(model="gpt-4-turbo", temperature=0)

@tool
def extract_cv_data(pdf_path: str) -> str:
    """Extract structured data from a candidate CV PDF."""
    with pdfplumber.open(pdf_path) as pdf:
        text = "
".join(page.extract_text() for page in pdf.pages)
    prompt = f"""Extract from this CV as JSON:
    {{"name","email","phone","skills":[],"experience_years",
     "education","last_role","companies":[]}}
    CV: {text[:3000]}"""
    return llm.invoke(prompt).content

@tool
def score_candidate(cv_data: str, job_description: str) -> str:
    """Score a candidate 0-100 against the job description."""
    prompt = f"""Score this candidate 0-100 for the role.
    Return JSON: {{"score": int, "strengths": [], "gaps": [], "recommendation": str}}
    JD: {job_description}
    CV: {cv_data}"""
    return llm.invoke(prompt).content

# Orchestrator
orchestrator = create_openai_tools_agent(
    llm, tools=[extract_cv_data, score_candidate, schedule_interview, send_email],
    prompt=orchestrator_prompt
)
agent_executor = AgentExecutor(agent=orchestrator, tools=[...], verbose=True)

Automation Pipeline

1. Ingestion: CVs uploaded as PDFs → pdfplumber extracts text
2. Screening: CV Screener agent extracts structured fields (skills, experience, education)
3. Scoring: Ranker agent compares each CV against the job description — outputs 0–100 score with reasoning
4. Shortlisting: Top N candidates above threshold automatically shortlisted
5. Scheduling: Scheduler agent checks Google Calendar API for interviewer availability → proposes slots
6. Communication: Emailer agent drafts personalised acceptance/rejection emails — HR approves before send

Results