Skip to main content
  1. Public practice questions
  2. AB-620 Practice Exams | AI Agent Builder Associate
  3. AB-620 Case Study Set

Question 1 — AB-620 Case Study Set

Paideffort Enterprise Leave and Workforce Agent
 

Business Requirements

Paideffort is deploying an enterprise HR agent to approximately 28,000 employees through Microsoft Teams and a corporate web channel. Employees must be able to submit leave requests by providing a leave type, start date, end date, and reason. The HR program office requires these four related values to be collected in one conversational interaction whenever the channel supports the required Adaptive Card capabilities. A separate confirmation can occur afterward.

Leave categories vary by employee region and are returned by the HR policy service at runtime. Start and end dates must be supplied before a request can be created. If an employee dismisses an input form, the workflow must not call the leave-creation service with incomplete data. The design must also distinguish responses from the original request card, a retry card, and a later confirmation card.

Paideffort requires Microsoft Entra ID-based authentication for production service-to-service access and prohibits application secrets from being embedded in topic definitions. The solution must support regional scaling, centralized audit records, and a future governance dashboard that correlates conversational submissions with downstream tool operations. Routine leave submissions should minimize unnecessary conversational turns and avoid avoidable model calls.

Technical Architecture

Microsoft Copilot Studio hosts the employee-facing topic. A Microsoft Foundry project contains an HR policy agent used when leave eligibility or policy interpretation requires agent reasoning. The application team uses the Foundry project endpoint with the Foundry SDK for project-scoped agent and tool operations. A separate Azure OpenAI endpoint is available to the OpenAI SDK for direct model operations that don't require project-scoped agent capabilities.

The Foundry agent includes a FunctionTool for deterministic entitlement calculations and an OpenAPI tool describing the HR service. An Azure Function implements the final leave-record creation operation. Production identities use Microsoft Entra ID and managed identity where supported.

Copilot Studio Topic Implementation

The topic retrieves the employee's valid regional leave categories into Topic.AvailableLeaveTypes. The table currently returned by the policy-processing step contains title and value columns. The leave form is then displayed by using Ask with Adaptive Card. The implementation team retained short default-style identifiers while prototyping the form.

{
  "type": "AdaptiveCard",
  "version": "1.5",
  "body": [
    {
      "type": "Input.ChoiceSet",
      "id": "input1",
      "label": "Leave type",
      "choices": "=Topic.AvailableLeaveTypes"
    },
    {
      "type": "Input.Date",
      "id": "input2",
      "label": "Start date",
      "isRequired": true
    },
    {
      "type": "Input.Date",
      "id": "input3",
      "label": "End date"
    },
    {
      "type": "Input.Text",
      "id": "input4",
      "label": "Reason",
      "isMultiline": true
    }
  ],
  "actions": [
    {
      "type": "Action.Submit",
      "title": "Submit request"
    }
  ]
}

The retry form and the later confirmation card also use Action.Submit. Operations wants responses to remain attributable to the specific card and action that produced them, including when an employee returns to an older card in the conversation.

trigger: LeaveRequest
steps:
  - getRegionalLeaveTypes:
      output: Topic.AvailableLeaveTypes
  - askWithAdaptiveCard:
      card: leaveRequestCard
  - callTool:
      name: CreateLeaveRequest
      inputs:
        leaveType: Topic.input1
        startDate: Topic.input2
        endDate: Topic.input3
        reason: Topic.input4
  - askWithAdaptiveCard:
      card: confirmationCard

A separate prototype transforms HR leave categories before the card is displayed. The source API returns records containing categoryName and categoryCode.

def prepare_leave_types(records):
    choices = []
    for record in records:
        choices.append({
            "name": record["categoryName"],
            "code": record["categoryCode"]
        })
    return choices

The pilot also contains a regional fallback path that asks four separate Question nodes for leave type, start date, end date, and reason before invoking the same leave-creation service. Employees using that path generate substantially more conversational turns than employees receiving the interactive form.

Future Requirements

Paideffort plans to add country-specific leave policies, additional Foundry agent capabilities, stricter Entra-based authorization, and centralized tracing across Copilot Studio, agent operations, and Azure Functions. The architecture must support higher transaction volume without multiplying unnecessary conversation turns or model invocations. Governance teams also require reliable variable names and action identifiers so that operational records can be correlated with the exact user submission that initiated each workflow.