AI Governance Studio: Find, Register and Govern AI Apps

Prukalpa Sankar

0:00 As we've spoken with customers about AI governance, we keep hearing three themes. The first, we've gone from shadow IT to shadow AI. Customers routinely tell us, "I don't even know how many AI apps and models or agents exist in my company. There could be hundreds, maybe there are even thousands." Then they tell us about AI regulation rapidly evolving. "It's changing every week. It's different at the federal level and the state level. I can't keep up." And then finally, they tell us how AI governance depends on data governance. How can you monitor an AI app if you don't know whether the underlying data has changed or not?

0:47 And so I'm gonna stop talking and hand it over to Tom to walk us through a demo of our AI Governance Studio. Tom, over to you.

Tom

0:56 All right. Thanks, PK. Let me switch screens for us all. Perfect. Get some of your faces out of the way here. Beautiful. All right. Organizations we are working with are wrestling the same things. I have thousands of developers building AI apps. How do I keep track of what they're building? Which models are available for use? How do I capture the context I need on a new app to be able not just to approve or reject, but truly advise on the best path forward? These aren't small problems, they're the foundation of responsible AI adoption, and they're exactly why we've built our AI governance module. So let's imagine I'm building a new AI application.

1:45 Before I start coding, I wanna know, has anyone else in the company already built something similar? With Atlan, that's no longer a blind spot. I can open the AI app homepage and instantly see every AI application across the company in one place. Now, I know what's already been built before I reinvent the wheel. So great, I've checked what's been built. The next question is: What models am I actually allowed to use for my application? Most companies pay for multiple models from multiple providers, and keeping track is nearly impossible. With Atlan, that confusion disappears. Models are automatically cataloged from my Databricks environment so I can search, browse, and understand every model available to me.

2:31 I've checked the apps. I've found the right model. Now it's time to register my AI application as an idea to be approved. For this demo, I'll use the UI so you can see the experience visually, but just remember, Atlan is an API first company with a robust SDK. Most enterprises automate this process with their CICD pipelines, so developers don't need to fill out forms manually. Here, you're seeing me fill out the basic details of the application. Then I associate the AI model powering it. In this case, I'll select Atlan AI. Next, I fill out the ethical AI section, documenting compliance requirements like the EU AI Act.

3:17 I can also capture any company specific standards here, like our internal fairness or transparency guidelines, so everything lives in one place. Once complete, the form routes to our governance team for review. Here's where they can request additional details, like whether the app is internal or external, which regions it'll be deployed in, and a few other details. I submit my app and move on to the next question: How do approvers know if they can approve an AI application? I heard from one customer that their legal team took three weeks to approve their chatbot because it's a new motion for them.

3:57 With our app framework, we've given you the ability to build agents to handle the velocity of change. Our launch partner, Tavro, built an AI risk classification agent that you're seeing here. The moment an application is submitted, the agent automatically classifies the risk associated with it, giving approvers clear guidance before making a decision. You'll see more about this later. Finally, what about monitoring? Most modern AI applications are RAG-based. With Atlan, I can visualize the lineage for every model from data sources to knowledge bases to downstream apps. That means I can continuously monitor compliance and ensure nothing slips through the cracks. AI is moving fast.

4:42 Atlan gives you the visibility and control to move with it responsibly and at scale. PK, turn it back over to you.

Prukalpa Sankar

4:56 As we built our AI Governance Studios, we realized that AI governance cannot be one size fits all. Regulations differ across regions, countries, even states, and it's also going to be constantly changing. And so we need an open, extensible ecosystem where customers and partners can create agents and apps to keep pace with this evolving AI regulation. I'm super excited to spotlight a partner, Tavro, that has built an AI governance agent that Tom just showed you about.

Tavro presenter

5:37 At Tavro, we are on a mission to help organizations unlock the true value of their data. Every enterprise today is facing AI sprawl. Hundreds of undocumented AI models and applications are being built and deployed, often with no clear monitoring in place. Take the example of a Fortune fifty financial services company. They had hundreds of AI use cases across the organization. But it could take six to eight hours to assess each use case to determine whether it is compliant with the critical regulations like EU AI Act. And there was no single place to automatically register and classify every use case across the enterprise.

6:26 To solve this, we built an agent that auto-classifies AI models and apps registered in Atlan's AI governance module. The Tavro agent analyzes the AI application metadata and classifies it as per the EU AI Act. The classification details as per Article five and six are captured as custom metadata within Atlan, along with the rationale behind the classification. This streamlined the AI application intake process, instantly tagging each application, and the time to complete the full risk classification was dropped to just minutes. The best part is that for end users, it is a unified and seamless experience.

7:12 The agent works natively within Atlan's AI governance module, essentially acting as an invisible copilot. For our developers, the build process was just as smooth. We were able to take what we had already created outside Atlan and integrate it with minimal reconfiguration. The app framework gave us powerful toolkits like temporal integration, which made monitoring the workflow far more efficient. And this is just the start. We are excited to continue building AI native apps on Atlan.

Prukalpa Sankar

7:50 Hopefully that gives you a sense of the kind of ecosystem we're envisaging, an ecosystem where we can open up innovation and allow our customers to stay ahead of the curve and not keep catching up to this ever-changing world.