All case studies / Case study 04 / 12
Corvane Services
Multi-agent onboarding orchestrator
AI agents that get everything ready for each new hire
In short
One message from HR is enough: AI agents split the work between HR, IT, equipment and communication, create the accounts, open the tickets and send the welcome messages, with human approval on sensitive steps.
- Client
- Corvane Services
- Sector
- Shared services centre
- Size
- 900 staff, ~40 hires a month
- 2 hinstead of 3 days to prepare an arrival
- 14 tasksautomated per hire
- 0missing account or laptop on day one
The context
With around forty arrivals a month, onboarding kept HR, IT and facilities busy with emails and spreadsheets. Some newcomers waited days for their laptop or their access.
The challenge
- 01
HR requests come as free text, rarely in the same format.
- 02
Each arrival triggers about ten tasks across four departments.
- 03
Some actions, like access rights or expensive equipment, must be approved by a person.
- 04
There must be a clear trail of who did what, and when.
Our approach
An orchestrator agent reads the request, extracts what matters (role, team, date, site) and plans the tasks. It delegates to specialised agents that act through tools connected with the Model Context Protocol: directory, Teams, IT ticketing, equipment ordering, email. A dashboard tracks each arrival and submits sensitive steps for approval.
The solution, step by step
- 01
The HR request
“Sarra joins the payroll team on 3 March in Tunis”: one plain message is enough.
- 02
Understood and planned
The orchestrator extracts role, team, date and site, then lists the tasks.
- 03
Split between agents
HR, IT, equipment and communication agents work in parallel, each with its tools.
- 04
Approved by a person
Sensitive access and purchases wait for the manager’s click.
- 05
Ready for day one
Accounts, laptop, badge and welcome message: everything is ready, and everything is logged.
Results
- −90%admin time per arrival
- D−2everything ready two days before arrival
- 100%of actions logged and auditable
- 4 departmentscoordinated without email ping-pong
What we deliver
- Orchestrator and specialised agents (LangGraph)
- MCP servers for the directory, Teams, IT ticketing and email
- Human approval rules
- React dashboard to track arrivals
- Full audit log
Technologies
- LangGraph
- Python
- TypeScript
- Model Context Protocol
- React
- PostgreSQL
Timeline
8 to 12 weeks
Team
2 AI engineers, 1 full-stack developer
Illustrative cases: clients and figures are fictional, representative of the solutions we deliver.