Welcome to Knowledge Graphs for AI Agent API Discovery, taught by Pavithra G K, who is head of business knowledge graphs, and Lars Heling, who is senior knowledge engineer at SAP Business AI. Large businesses may have thousands of APIs that have been made available by many different teams to enable other teams to carry out many possible tasks across the business. For example, APIs to help buy or return an item or request travel approval or request IT approval for accessing data, and so on. I think it's great that people are building so many APIs for AI agents, but this actually makes it hard for AI agents to figure out which APIs to use and in what sequence to carry out a complex business process. In this course, you'll learn how to solve this problem using knowledge graphs. To help an agent use APIs, you bring raw API specifications into a knowledge graph. This gives a structured view of all the APIs. Additionally, you add information to the graph to tell it what is the right order in which you should call the APIs. Take the example of an agent given the task of purchasing an item. It might search for and find the purchase order API as the most relevant API and maybe want to call that API to place the order. But here, the agent might be unaware that in a big company, it first needs to call another API to create a purchase request, which is then to be approved by the IT department before finally it places the actual order. This processed information that the steps needed are to first generate a purchase request, then get IT approval, and then finally call the purchase order API. This information may not be available in the definition of the APIs. And without this type of information, your knowledge graph and your agent's ability to understand this important business process workflow and execute it will be incomplete. That's right, Andrew. To capture this, you will extend the knowledge graph with process data. This turns your graph into not just a list of APIs, but a network of how they are connected in real processes. The agent is now aware that the purchase request API needs to be called first before calling the API to create an order. To give the right APIs to the agent to perform a given task, you will start with a semantic retrieval. You will embed all the APIs and their specifications, and use similarity search to find the best matches. For example, if the agent wants to create a purchase order, it will retrieve semantically related APIs like the purchase order API. Calling this API directly will not be the right action, as the business process might require the purchase request API to be called first for the approval of the purchase by a manager. To solve this, you will extend the search on the knowledge graph to find and include related APIs that the business process requires. This will give the agent all the necessary APIs and the order in which they should be called. In the last lesson, you will put all of this together to build an AI agent that can carry out real tasks while following the correct process sequence. Many people have worked to create this course. I'd like to thank Christoph Meyer, Felix Sasaki, and Johannes Hoffart from the SAP team. From DeepLearning.AI, Esmaeil Gargari also contributed to this course. Businesses have complex and important processes to get work done. Implementing agentic workflows to carry out these processes is a huge part of the work that lies ahead for the AI community. And getting this right is an important problem. The first lesson will be an introduction to knowledge graphs, what they are and how to use them to help agents understand and follow complex multi-step processes. So, let's go to the next video and get started.
Knowledge Graphs for AI Agent: API Discovery