Loading Now

Compliance Academy: A Multi-Agent Cyber Mystery Built on Microsoft Foundry Agent Service

Join the Live Coding Battle at Microsoft Reactor

On 10th June at 9 am PT, Microsoft Reactor is hosting an exciting live coding battle. Three talented individuals will be competing to demonstrate the capabilities of reasoning agents in real time before a streaming audience.

<h2>Setting the Stage for Creativity</h2>
<p>While a straightforward Q&A or a document summariser might seem like an easy choice, I opted for a more ambitious project. I’m introducing **Compliance Academy**: a captivating role-playing game where you'll assume the role of the chief investigator at <strong>Helix Dynamics</strong>, a fictional biotech firm that has lost 14 GB of critical clinical trial data. In this game, you'll interrogate five suspects to uncover a single perpetrator while navigating various compliance frameworks such as SOC 2, HIPAA, and ISO 27001.</p>

<h3>Why Compliance Academy?</h3>
<p>Most professionals view compliance training as a dull task they rush through just to meet a deadline. The information is usually dense, and the stakes may seem abstract. Many just learn enough to pass a quiz, which isn't the right approach for material that should genuinely impact decision-making.</p>
<p>With reasoning agents, we can flip this notion around. Instead of just presenting information, a multi-agent system creates scenarios where learners take charge. They are encouraged to ask questions, interrogate evasive suspects, and compile evidence. When a <strong>Compliance Officer</strong> agent later steps in to reference SOC 2 CC6.1 and Helix Dynamics' own policies, it resonates because the player has truly engaged with the material.</p>

<h2>How Does It Work?</h2>
<p>Built on the <strong>Microsoft Foundry Agent Service</strong>, and utilizing <strong>Azure OpenAI</strong> and <strong>Azure AI Search</strong>, Compliance Academy is designed for immersive learning. Here’s a brief overview of how the system functions:</p>
<ul>
    <li><strong>Connected Agents Pattern:</strong> There are always four core agents providing expertise, while additional suspect agents activate dynamically throughout the investigation.</li>
    <li><strong>Deep Personalities:</strong> Each suspect agent has unique backstories, alibis, and personalities. You'll encounter characters like an HR Director, an IT Admin, and a summer intern, each with hidden truths waiting to be uncovered.</li>
    <li><strong>Interactive UI:</strong> The audience interacts with a user-friendly Chainlit interface, featuring action buttons for briefing, suspect interrogation, and evidence gathering.</li>
</ul>

<h3>One of the Most Innovative Features</h3>
<p>The <strong>Scenario Generator</strong> is one of the standout features of this project. During the demo, participants can propose a breach scenario in just a sentence. Within about forty seconds, a fresh scenario loads into the game, allowing players to explore new challenges instantly.</p>

<h2>Key Takeaways for Future Projects</h2>
<p>Here are some essential lessons I learned that will guide my future projects involving reasoning agents:</p>
<ol>
    <li><strong>Ground Compliance in Retrieval:</strong> Always base compliance-related content on source material to avoid misinformation.</li>
    <li><strong>Prioritise User Experience:</strong> Observability isn’t an afterthought; it enriches the experience, especially during a live demo.</li>
    <li><strong>Leverage Corrective Feedback:</strong> Allow the model to self-correct its outputs, ensuring a smoother user interaction.</li>
    <li><strong>Specialise Agents:</strong> Each agent should have a distinct role to ensure efficiency and clarity.</li>
    <li><strong>Have Backup Plans:</strong> Ensure that you have a terminal-driven backup ready in case of unexpected issues.</li>
</ol>

<h2>Why Use Microsoft Foundry Agent Service?</h2>
<p>This platform is particularly well-suited for this project as it aligns perfectly with the Connected Agents framework. It offers seamless integration, making it easy to deploy and manage in one resource group.</p>

<h2>Join Us Live!</h2>
<p>Don’t miss out on witnessing how we tackle the Agents League Challenges! Watch live at <a href="https://developer.microsoft.com/en-us/reactor/events/26942/">this link</a>.</p>

<h2>Getting Started with the Project</h2>
<p>If you’re interested in diving deeper, the repository is open source! You can clone it and follow the instructions below:</p>

<pre>
git clone https://github.com/lwhieldon/msft-enterprise-learning-agent.git
cd msft-enterprise-learning-agent
python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt
chainlit run app.py -w
</pre>

<h2>FAQs</h2>
<h3>What is Compliance Academy?</h3>
<p>It’s a role-playing game where players investigate a data breach, interrogating suspects and learning about compliance through interactive scenarios.</p>

<h3>How do I participate in the event?</h3>
<p>You can join the live broadcast on 10th June at 9 am PT via the Microsoft Reactor link provided.</p>

<h3>Is the source code available?</h3>
<p>Yes! The entire codebase is open-source and available for cloning and further exploration.</p>

<h3>Who is the author of this project?</h3>
<p>Lee Whieldon, a Principal at SC&H Group, leads the Data Analytics & AI advisory practice, focusing on structured data and reasoning agents.</p>

Share this content:


Discover more from Qureshi

Subscribe to get the latest posts sent to your email.

Discover more from Qureshi

Subscribe now to keep reading and get access to the full archive.

Continue reading