Welcome to this course on Nvidia's NeMo Agent Toolkit, which will show you how to harden your agent for production. This course is built in partnership with Nvidia and is taught by Brian McBrayer, who is Solutions Architect in Generative AI at Nvidia. One problem that many of us face is how to take that demo that works maybe 60% of the time and turn into something reliable that you can safely deploy to users. Nvidia's open source NeMo Agent Toolkit or NAT gives key tools for doing so. Specifically, NAT makes it easy for you to add observability to build and repeatedly run evals and also deploy your agentic workflow. Maybe you've already built an agentic workflow in raw Python or using a framework like LangGraph or Crew AI. You can then integrate NAT to get observability to see what your agent workflow is doing. For example, if your agent occasionally calls the wrong tool, NAT will show you execution traces so you can actually see what's going on. In addition, NAT makes it easy for you to set up and run evals in a disciplined way. This is very helpful if you want to run maybe a CI/CD process or if you just want to drive a systematic process for improving your agent's performance. In this course, you will learn NAT by building an agent to analyze climate data. You'll start by registering Python functions as tools to add more functionality to your agent. Agents reason in natural language, so without the right tools, it's hard to know why they made certain decisions. You'll use OpenTelemetry tracing, visualized in Phoenix, to see exactly how your agent reasons, which tools it considers and what choices it makes. You'll also evaluate performance over time, so when you adjust a prompt or change a model, you'll know whether it actually helped. NAT works directly with frameworks you already use, LangChain, Crew AI, or your own custom code. So, finally, you'll expand the climate analyzer agent to a multi-agent workflow where specialized agents built with frameworks like LangGraph collaborate on different aspects of climate analysis. Deploying agents introduces so many complexities. You need an API endpoint, authentication, rate limiting. Suddenly you're writing infrastructure code instead of working on your agents. And all of these problems reappear anytime you want to iterate on the models you're using, or the prompts you're passing in. Through several labs in the course, you will learn how NAT makes AI agents ready for production. Many people have worked to create this course. I'd like to thank from NVIDIA, Josh Wyatt, Michael Demoret, and Bartley Richardson. And from DeepLearning.AI, Summer Rae, Esmaeil Gargari, and Brendan Brown. In the first lesson, you build your first agent in NAT, connect it to an API endpoint and see it working in real time. Let's go on to the next video to get started.