Meet Onboard_D, your HR onboading coach!

This solution has been made as part of the Microsoft AI Agents Hackathon held in April 2025.

Golnoosh Ameri

Igor Bertnyk

Franck Cornu

Mohamed Iftiquar Aslam Hameed

Achinta Sarkar

What is it?

Onboarding processes can sometimes be overwhelming for new employees, with numerous tasks to complete, trainings to attend, and people to meet. Not all companies have a clear onboarding process or a dedicated coach to guide new hires effectively. To ensure employees are not left to navigate this alone and to provide a great experience, we created Onboard_D. This dedicated AI assistant helps new employees get started by answering frequently asked questions and proactively notifying them of tasks they need to complete according to a specific onboarding plan set by the HR department.

Don't do this at home!

This solution is not suitable for production! It has been made in less than 3 weeks for the sole purpose of this Hackathon

It includes the following features:

• Provide assistance with HR-related inquiries: Address a wide range of questions that employees may have, including policies, benefits, procedures, and other human resources topics.
• Summarize the current tasks assigned to employees: Offer clear and concise summaries of tasks or responsibilities assigned to individuals, ensuring they have a comprehensive understanding of their workload.
• Proactively notify employees about tasks and critical attention points in their onboarding plan: Deliver timely reminders and updates regarding assigned tasks, deadlines, and important milestones to guide employees through the onboarding process effectively.
• Offer a customizable solution for administrators: Provide administrators with the flexibility to tailor systems or functionalities to meet their specific needs and organizational requirements.

Here are some conversations examples with Onboard_D (click on the image to get the full view):

• 
  Get current assigned tasks for user
• 
  Get details about a specific task from the list
• 
  Update task status with a "Human-in-the-loop" step

What concepts it showcases?

Creating agents that tell jokes or generate cat images can be fun but not that useful in your work day-to-day life. We wanted to create something meaningfull and reusable enough to be used by any company on top of Microsoft 365 leveraging AI where it is good at: understanding intents, analyzing amd summarizing stuff. Therefore, the Onboard_D agent showcases the following:

• Integration with Microsoft 365 as part of an end-to-end solution, leveraging Teams, Planner and SharePoint.
• Sending proactive agent nofications to users.
• Handling agent authentication with both delegated (SSO) and application permissions (proactive notifiations) whithin Microsoft Teams and the Microsoft Graph API.
• Handling multiple tools whithin an agent performing different set of actions.
• Handling agent tool outputs and artifacts to display data in adaptive cards.
• Handling Human-In-The-Loop pattern in pair with the Bot Framework for sensitive actions, like POST requests initiated by tools.

info

For the purpose of this hackathon (and also because of time!), we only implemented a small set of features.

How this works?

Here is the overal architecture of the solution with the key components:

The solution is split in three main parts:

Setup the employee plan

Before interacting with the agent, the onboarding plan for the new employee needs to be created. The solutions uses a Microsoft 365 group for HR staff members to manage new employee hires. The flow is as follow:

1. First, HR staff create a new Planner plan in the group that will act as default oboarding plan template for all employees.
2. When a new employee is hired, HR staff add a new entry in a dedicated SharePoint employees list with the planned start date.
3. A Power Automate is automatically triggered performing the following actions:
   • Get the onboarding template and and duplicate it for the new employee with tasks assigned to him.
   • Wait until the employee start date is reached. When reached, sends notification to the agent every X days (10 minutes in that demo) for that specific employee.

Index HR content

This part is aimed to index relevant documents so employees can ask general HR related questions to the agent (like policies). The index is also used to retrieve specific documents associated to tasks as references (for instance documents explaning a task in depth beyond the traditionnal description field).

1. HR Staff uploads relevant document in a dedicated SharePoint library.
2. Optionally, they can add the document as reference for a particular tasks in the template, giving employee material to clearly understand the task.
3. When a document is updated in the library, a Logic App is triggered, indexing the document in Azure AI Search, using Azure Open AI to generate vector embeddings.

Interact with the agent

Last but not least, employee can engage directly with the agent through Teams by asking questions about its current tasks created in its onboarding plan, or get details about specific task. He can also ask general question related to HR. Depending on the request, the agents uses multiple tools to communicate with the different systems (Planner, SharePoint and Azure AI Search). For more sensitive tasks, like updating the status of a particular task (ex: complete it), the agent will ask for an explicit confirmation first ("human in the loop").

The agent will also notifies the user and remind him his current task every X days ('X' being configurable).

What technologies are used?

This solution is built and/or involves the following technologies:

Technology	Used for
Azure OpenAI	LLM model to answer user questions. We use the gpt-4o-mini model for agent (better suited for chat completions) and text-embedding-3-large for embeddings both from OpenAI.
Azure AI Search	Indexing HR documents from SharePoint
Microsoft Planner	Onboarding plan definiton for both template and users.
SharePoint	Store HR documents in a library adn mange employes onboarding list.
Microsoft Bot Framework	User interactions whithin Microsoft Teams handling dialogs and SSO authentication.
Langchain JS	LLM Orchestration with the Bot Framework (tools definition and calls, "human-in-the-loop")
Power Automate	Employee plan setup and agent notifications.
Azure Logic App	HR documents indexing logic.

Test it!

Now you now what it is and how this works, let test it! First you'll have to setup your environment.