Building AI Agents with ADK and MongoDB Vector Search

"Building an AI application that can act on real data involves more than connecting an LLM to a chat interface. The model needs access to up-to-date information, a way to retrieve data when a user’s language does not match the stored fields and tools to perform actions. It also needs to decide which tools to call and in what order for each request. In this hands-on workshop, you’ll address these challenges by building a grocery shopping assistant with Google’s Agent Development Kit (ADK) and MongoDB Vector Search. You’ll see how an AI agent can interpret natural-language requests, find relevant products through semantic search, coordinate multiple tool calls and act on product and cart data. You’ll begin by developing an intuition for the differences between simple prompting, retrieval-augmented generation (RAG) and AI agents, including when an agent is the right choice for a problem. Through a series of guided exercises, you’ll turn a basic LLM-powered application into a multi-tool agent. The agent will understand natural-language requests, choose the right actions and use the results to complete each task. Your grocery shopping agent will learn to use three tools: Find products by meaning, not just exact keywords, using semantic search powered by MongoDB Vector Search. Add selected products to a shopping cart. Calculate the cart total. Along the way, you’ll explore how vector embeddings and search indexes make product discovery more flexible. For example, it can help the agent interpret requests such as “sweet treats” or “ingredients for a chocolate cake.” You’ll also see how tools are exposed to an agent, how the agent chooses and sequences tool calls, and how MongoDB supports practical AI applications that work with real data. The workshop combines brief explanations with hands-on practice. You’ll work with a real showcase application, experiment with prompts like “Add milk and flour to my cart” and “What’s the total cost of my cart?”, inspect the agent’s tool executions and discuss how these patterns can transfer to your own projects. You can also try a multimodal prompt using an image alongside text. At the end of the session, you’ll have the opportunity to complete a skill check and earn the MongoDB AI Agents Skill Badge, a shareable Credly credential based on the concepts covered in the workshop."