OnboardIQ

Inspiration

Every time a company hires someone, HR manually completes 8+ tasks across 5+ tools — sending welcome emails, raising IT tickets, dispatching compliance forms, briefing the manager, and scheduling meetings. This takes 2-3 hours per hire, is error-prone, and often leaves new employees without access on Day 1. We wanted to eliminate this entirely using AI agents.

What it does

OnboardIQ is a fully autonomous multi-agent onboarding orchestrator built on Airia. When HR confirms a new hire, a single trigger activates an intelligent 6-node workflow:

OnboardIQ Orchestrator analyses the hire (role, region, department) and generates a complete onboarding brief
Compliance Validator checks regional requirements across 7 countries (India, USA, UK, UAE, Canada, Australia, Singapore)
HR approves via Human-in-the-Loop checkpoint — one click
Four sub-agents execute in parallel:
- WelcomeMessenger: Personalised welcome email + manager brief
- ITProvisioner: Role-specific IT provisioning ticket
- DocumentDispatcher: Correct compliance forms by region
- CalendarScheduler: Day 1 orientation + Day 2 manager 1:1
Completion Summarizer generates audit report and writes structured nodes to Apache AGE Knowledge Graph

How we built it

Built entirely on the Airia platform using:

Multi-agent orchestration with nested sub-agents via MCP Tool interfaces
Human Approval (HITL) node from Airia's Flow Control
Apache AGE Knowledge Graph for persistent audit trails
RAG-powered data sources (onboarding rules, provisioning maps, email templates, compliance forms)
GPT-4o for the Orchestrator's analysis node
GPT-4o mini for all 4 sub-agents (cost efficiency)
Graph Database Cypher Query tool for structured memory

Challenges we ran into

Wiring the MCP Tool interfaces correctly so sub-agents are callable from the Orchestrator
Configuring the Knowledge Graph Cypher queries to write multiple node types without duplicates (used MERGE instead of CREATE)
Passing variables correctly across nested agent calls
Balancing model choice — GPT-4o for reasoning, GPT-4o mini for structured tasks

Accomplishments that we're proud of

Full end-to-end working system: one input triggers 5 agents
HITL checkpoint that genuinely pauses the workflow pending human approval
Knowledge Graph that auto-populates Employee → Manager → Task nodes with labelled relationships after every onboarding
Region-aware compliance covering 7 countries with zero code changes
Cost under $0.01 per complete onboarding run

What we learned

Airia's MCP Tool interface is incredibly powerful for agent-to-agent communication
Knowledge Graphs add a layer of enterprise credibility — audit trails matter in HR
HITL design is what separates a demo from a responsible enterprise product
RAG retrieval from structured knowledge files is more reliable than stuffing everything into prompts