Efficient Query Routing using Agentic RAG

The industry is abuzz with the term "Vertical Agents," yet there’s often little clarity on how Agents work. At its core, however, an Agent is simply a workflow designed to automate tasks by using environment tools and enhancing its ability to reason, plan, decompose, and execute a given task.

In this talk, we’ll explore the necessity of planning and reasoning in various industries and build a practical use case in the Education domain. The application will answer user questions based on provided academic notes, and if the information isn’t available, it will seamlessly search the web. The decision-making process—whether to use existing knowledge or external sources—is where an Agent's role shines. To define and manage a custom knowledge base, we’ll utilize a Retrieval-Augmented Generation (RAG) approach.

With Agentic RAG, the audience will gain clarity on routing concepts, learn how to design an efficient architecture and understand how the Thought-Action-Observation loop enables an Agent to function effectively. As for the technical aspect, we will use LlamaIndex for the routing with its ReAcT Agent and Qdrant as the vector database with Hybrid Search capabilities.