NVIDIA at Microsoft Build | Builder’s Arcade

stacks of vertical green yellow and blue bricks

Thank you for joining NVIDIA at Microsoft Build

row of colorful abstract blocks on black background

Thank you for joining NVIDIA at Microsoft Build

NVIDIA Builder’s Arcade

The NVIDIA Builder’s Arcade offered participants five days of hands-on activities designed to accelerate developers’ ability to build, deploy, and scale agentic AI with NVIDIA solutions on Microsoft Azure.

Missed the developer challenges?

Although the Builder’s Arcade challenge is over, the featured activities are still available for you to explore on your own. You can also take advantage of the opportunity to check out NVIDIA solutions for building AI in the NVIDIA NGC Catalog.

Explore the catalog

Start

Run your first AI workload

You’re an enterprise developer who’s been tasked with evaluating NVIDIA AI on Azure. Get a feel for how AI workloads run on Azure with one-click inference using an NVIDIA accelerated-powered endpoint.

Level Up:

Your challenge begins now—check the instructions on the right to make your first move.

Instructions:

Step 1: Watch the City of Raleigh case study
Learn how a real city uses NVIDIA AI on Azure for safety and mobility. Watch the video to understand the real-world application of the inference workload you’re about to run.
Step 2: Demo agentic video search
Learn how to use the new VSS skills with coding agents: scroll down to the video in the linked technical blog to demo agentic search by attributes, events, and actions.
Step 3: Sign up for NVIDIA’s NGC Catalog
Create your NGC Catalog account. You’ll need this to access and connect NVIDIA NIM (inference microservices) throughout the 5-day journey.
Step 4: Run the one-click inference demo
Using the build.nvidia.com link above, click “View Examples,” and select “Warehouse with Object Tracking.” Run the demo and observe how the NVIDIA-accelerated endpoint processes the video feed in real time.

Deploy

Turn models into usable services

Your team needs a lightweight, high-performance language model deployed as an API. Lock into the process of models becoming APIs when you launch Nemotron Nano 3 as an API endpoint.

Level Up:

Your challenge begins now—check the instructions on the right to make your first move.

Instructions:

Step 1: Explore Nemotron Nano 3 on Microsoft Foundry
Using the link above to view Nemotron Nano 3 in the Microsoft Foundry model catalog, review the model card: What is it optimized for? What are the deployment options?
Step 2: Walk through deployment configuration
Select the model and review the deployment configuration steps. Observe how NVIDIA NIM packages the model as a ready-to-deploy microservice endpoint.
Step 3: Watch “Accelerating AI model performance”
View the Microsoft Reactor session on accelerating AI model performance (AMER). Take notes on how this connects to what you just explored in Microsoft Foundry.

Optimize

Improve performance and efficiency

Your agentic pipeline from Day 3 is running, but it’s too slow and expensive for production. Experiment with GPU configurations—adjust settings, simulate different scaling scenarios, and document the trade-offs you find between latency, throughput, and cost.

Level Up:

Your challenge begins now—check the instructions on the right to make your first move.

Instructions:

Step 1: Watch “Getting More from the GPUs You Already Have”
View the BrightTALK session on GPU optimization. Take notes on the key levers: batch size, model parallelism, quantization, and scaling strategies.
Step 2: Explore the NVIDIA NeMo Agent Toolkit
Review the NeMo Agent Toolkit. Understand how it helps builders create, optimize, and evaluate AI agents—including performance tuning for production workloads.
Step 3: Review NVIDIA Control Panel settings
Explore the NVIDIA Control Panel to understand GPU configuration options. Even if you’re not adjusting settings live, familiarize yourself with what’s available for performance tuning.

Operate

Design for real-world deployment

You’ve built, deployed, and optimized an AI pipeline. Now, close out your journey with a production-ready perspective by using NVIDIA’s Blueprint tools to generate a system architecture for your AI application.

Level Up:

Your challenge begins now—check the instructions on the right to make your first move.

Instructions:

Step 1: Explore NVIDIA AI Blueprints
Visit the link above to review the available reference architectures. Identify which blueprint aligns most closely with the pipeline you built on Days 1–4.
Step 2: Explore the Video Search and Summarization Blueprint
Walk through the Video Search and Summarization reference architecture. Note how it structures compute, retrieval, and generation for a production use case.
Step 3: Generate your architecture blueprint
Using what you’ve learned across all 5 days, sketch or generate a system architecture blueprint for your AI application. Include: model deployment (Day 2), agentic pipeline (Day 3), GPU optimization strategy (Day 4), and production infrastructure.

Build

Connect capabilities into workflows

You’re designing an AI agent that can pull enterprise data, reason over it, and generate a structured output. Get comfortable with chaining NVIDIA NIM into agentic workflows by building a multistep agentic pipeline.

Level Up:

Your challenge begins now—check the instructions on the right to make your first move.

Instructions:

Step 1: Watch the Deep Research Agent session
View “Build a Deep Research Agent for Enterprise Data with Scalable Performance” on Microsoft Reactor. Note the architecture pattern: retrieval, reasoning, and generation as distinct pipeline stages.
Step 2: Watch “Developing with NVIDIA Nemotron on Microsoft Azure”
View the video linked above for a quick explanation of the benefits to developers of using NVIDIA Nemotron open-source reasoning models with Microsoft Azure.
Step 3: Explore the Nemotron models experience
Visit the above NVIDIA Nemotron page that lets you interact with Nemotron models directly. Explore the models and select one to test. Test a prompt and observe how the model handles multistep reasoning.
Step 4: Review the synthetic data generation use case
Explore NVIDIA’s Synthetic Data Generation for Agentic AI resource. Understand how synthetic data trains and evaluates the agents you’re building.
Step 5: Map your pipeline and complete
Sketch or outline your retrieve → reason → generate pipeline. Identify the NVIDIA NIM or tool at each stage.

horizontal stacksof colored blocks on black background

New from Microsoft and NVIDIA: Announcements from Microsoft Build

Together, Microsoft and NVIDIA deliver a unified stack for accelerating physical and agentic AI, from devices to the cloud. Discover new developments and integrations unveiled during Microsoft Build—read the article to learn more.

Read announcements

Replay the highlights

An event this big isn’t over just because the agenda ends. Revisit top moments and key takeaways when you view and share the recaps and highlights from these industry voices.

Santiago Valdarrama

Linda Haviv

Jess Ramos

Gray dashed lines and small clusters of colorful blocks

What next?

Ready for more opportunities to experiment with what’s next in AI-driven apps? Engage with developers live through Microsoft Reactor AI Apps & Agents Dev Days, a monthly series where ideas turn into working AI experiences. See upcoming events and explore past sessions on demand.

Learn more