Build with AI: Production-Ready AI Agent Systems on Databricks

Course details

Building AI agents is becoming more accessible, but moving from a working prototype to production requires knowing when multiagent architectures make sense, how to evaluate their performance, and which tools can help you build more reliably.

In this course, join Viktoria Semaan, Principal AI Evangelist at Databricks, as she demonstrates how to design and deploy a complete multiagent system in Databricks using a realistic bookstore scenario. Explore multiagent architecture patterns, evaluate agent performance with MLflow 3, and apply Databricks tools to accelerate development. Along the way, examine each system component in detail, troubleshoot common integration challenges, and explore reusable architecture patterns alongside working code from the course repository that you can adapt to your own unique use cases.

Concepts

Introduction

Building enterprise AI agents.
What this course covers

Data Foundations

AI functions for parsing data
Building the products table

Vector Seach

Understanding vector search
Creating and testing a vector index

Custom RAG Agent

RAG concepts and playground prototyping
Building custom RAG agent in notebook
MLflow and evaluations
Custom scorers and deployment

AI Genie

AI BI Genie - Generate tables
AI BI Genie - Setting up the space

Knowledge Assistant (No-Code RAG)

Creating a Knowledge Assistant

Evaluating Knowledge Assistants

Improving Knowledge Assistant with human feedback

MCP Servers

Adding external MCP (web search)

Supervisor Agents

Building and testing supervisor agent

Conclusion

Troubleshooting and clean-up