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

Question 41 — AB-620 Case Study Set

Paideffort Regulatory Intelligence Agent
 

Business Requirements

Paideffort operates a financial-services advisory platform used by compliance officers in 14 regions. A Microsoft Copilot Studio agent answers questions about investment communication standards, disclosure requirements, and internal approval procedures. Regulatory content changes frequently and is maintained in a proprietary regulatory API, so compliance leadership doesn't want static uploaded files to become the primary source for these answers.

Each response must be grounded in the most relevant regulatory information available for the officer's current question. Where available, answers should provide a meaningful citation label and a link to the corresponding regulatory source. The regulatory API can return between 8 and 50 matching records, but the compliance team wants the agent grounded only on the strongest results rather than a broad unranked catalog.

The initial integration retrieves data from one API with one request. The API response can be reshaped directly with Power Fx and doesn't require pagination or multi-system aggregation. A future release will combine the regulatory API with a policy database and regional exception service, then apply business rules before returning material to the topic.

Technical Architecture

Copilot Studio hosts the compliance topic and its Create generative answers node. A Microsoft Foundry project provides a separate reasoning agent for complex compliance investigations. Project-scoped agent operations use the Foundry project endpoint through the Foundry SDK, while direct model operations use an Azure OpenAI endpoint through the OpenAI SDK.

The Foundry agent has a FunctionTool for deterministic risk calculations, an OpenAPI tool for an approved case-management API, and access to an Azure Function that records review decisions. Production access uses Microsoft Entra ID and managed identities where supported. The regulatory retrieval used by the generative answers node remains deterministic and is executed before response generation.

Current Regulatory Retrieval

The officer's question is stored in Topic.ComplianceQuestion. The topic then calls the regulatory API and stores the JSON response in Topic.RegResponse. The current endpoint returns up to eight recently updated regulatory records.

topic: RegulatoryAnswer
steps:
  - captureQuestion:
      output: Topic.ComplianceQuestion

  - sendHttpRequest:
      method: GET
      url: "https://regulatory.paideffort.internal/regulations?limit=8"
      response:
        saveAs: Topic.RegResponse

  - setVariable:
      variable: Topic.RegData
      formula: MapRegulatoryResults(Topic.RegResponse.results)

  - createGenerativeAnswers:
      dataSources:
        mode: ClassicData
        customData: Topic.RegData

The API supports a query parameter that performs relevance ranking against the officer's question before results are returned. Without that parameter, results are primarily ordered by regulatory update time.

Response Transformation

Each API result contains regulationText, sourceUrl, regulationTitle, region, and lastUpdated. A developer created the following Power Fx transformation for the variable passed to Custom data:

ForAll(
    Topic.RegResponse.results,
    {
        ContentText: regulationText,
        ContentLocation: sourceUrl,
        Title: regulationTitle
    }
)

During pilot testing, the API returned eight records for a question about projected investment returns. The fourth result happened to contain the most directly applicable prohibition, while the first three were recently updated disclosure records. Compliance analysts expected the most applicable rule to influence the generated answer.

A platform engineer has proposed placing the existing single HTTP call inside a Power Automate flow immediately, even though the flow would perform no aggregation, pagination, connector operations, or additional business logic. Another proposal keeps the current HTTP request and Power Fx transformation until the planned multi-system regulatory release.

Planned Multi-System Release

In the next phase, Paideffort will combine regulations from the proprietary API, regional exceptions from a SQL-backed policy service, and internal approval guidance exposed through a Power Platform connector. The retrieval layer must remove superseded records, apply regional rules, rank the combined results, and return a compact table to Copilot Studio.

{
  "regulationText": "Projected return guarantees are prohibited in client communications.",
  "sourceUrl": "https://compliance.paideffort.internal/framework/article-7",
  "regulationTitle": "Financial Conduct Framework - Article 7",
  "region": "Global",
  "lastUpdated": "2026-08-29"
}

Governance and Future Requirements

Paideffort plans to expand the agent to additional jurisdictions while preserving traceable citations and minimizing irrelevant grounding content. Governance dashboards will record the user's query, selected source titles, source locations, retrieval method, and downstream review outcome. The architecture must support future scaling without sending large unfiltered result sets to the generative answers node or duplicating complex retrieval logic across multiple topics.