The best way to learn a new toolkit is to build something real, and that’s exactly what developers did at the recent NVIDIA NeMo Agent Toolkit Hackathon. Over two weeks, participants across skill levels—from students to seasoned professionals—experimented, prototyped, and created intelligent multi-agent AI workflows using the open-source NeMo Agent toolkit (formerly known as the AIQ toolkit).

With access to example projects, technical documentation, and office hours with NVIDIA engineers, participants explored the toolkit’s orchestration, memory, and profiling features—often combining them with other NVIDIA technologies and contributing improvements upstream. 

Submissions were judged on real-world applicability, technical execution, and how effectively they demonstrated agentic AI in action. The results were inspiring: functional, innovative projects that showcase how AI agents solve real problems across logistics, software development, and personal travel planning. 

First place: Route Optimization Agent for Intelligent Fleet Management

Developed by a member of the TCS Smart Mobility Group, the winning project demonstrated a powerful integration of the NVIDIA NeMo Agent toolkit, NVIDIA cuOpt, and NVIDIA Omniverse libraries to solve complex logistics and supply chain challenges. 

This intelligent system orchestrates a team of AI agents—each specializing in natural language understanding, constraint extraction, and route computation. The agents collaborate to interpret user instructions, such as “plan optimized routes for three forklifts to deliver material from storage to trucks, each with limited capacity,” extract operational constraints, and invoke cuOpt to compute optimal routing plans in seconds.

TCS developed an application on the Omniverse Kit SDK to provide a simulation environment, test scenarios, and train agents safely. Once optimized, the routes are deployed onto NVIDIA Jetson Nano-powered autonomous mobile robots (AMRs), such as MyAGV robots, completing a full simulation-to-deployment pipeline.

Key features:

• Natural language interface: Enables users to describe logistics tasks in plain English, enabling intuitive interaction without the need for technical commands.
• Dynamic multi-agent orchestration: Chains together language models, optimization engines, and data pipelines to automate complex tasks from input interpretation to execution.
• Real-time optimization: Utilizes cuOpt to generate rapid, constraint-aware route plans in dynamic real-world environments.
• Simulation-to-deployment pipeline: Uses Omniverse to simulate and validate routes before deploying them to physical robots, ensuring safe and effective real-world execution.

This project demonstrates how agentic AI—when paired with high-performance NVIDIA CUDA-X solvers like cuOpt—can drastically streamline fleet management, reduce operational costs, and enable faster, smarter operational decisions.

Second place: OpenCodeReview

OpenCodeReview empowers developers with automated, AI-driven code analysis to detect security vulnerabilities and improve code quality. Built using the NeMo Agent toolkit, the system scans selected files, highlights issues, and recommends fixes, integrated into existing workflows effectively.

A standout feature is the ability to swap between different AI models by adjusting configuration files, enabling customization across coding standards or team needs. Developers don’t need to prompt-tune anything—the agent instructions are pre-configured for ease of use. Developers can also flexibly use different AI models—some models may be better suited for certain programming languages, and more easily select the best models for their use case.

By leveraging the NeMo Agent toolkit’s orchestration and memory features, OpenCodeReview democratizes secure coding practices, making advanced review accessible to individual developers, startups, and large organizations alike. The toolkit also makes it easy to update or add additional AI agents by simply changing the configuration files.

Third place: AI-Powered Travel Planner and Experience Agent

This modular travel assistant, built directly into the NeMo Agent toolkit’s examples folder, demonstrates how agentic AI can unify fragmented travel tasks into a single, conversational experience. This submission included a large number of tools and an impressive number of features. For example, users can: 

• Search and book flights using natural language queries across APIs.
• Plan end-to-end journeys including hotels, activities, and maps.
• Rely on resilient data access, with a local fallback database in case of API outages.
• Interact naturally across tasks, with contextual memory preserving the flow, from flights to hotels to local recommendations.
• Visualize travel plans through embedded maps and destination data, a feature we didn’t even realize was possible with the UI prior to this submission.

The workflow is powered by the DeepSeek LLM to orchestrate queries and API interactions and is easily swappable with other models. This tool highlights the power of multi-step, memory-aware agentic interactions in personal productivity applications. 

Honorable mention: Cyber Agent for Detecting Very Small Indicators of Compromise

This cyber defense tool is designed to detect subtle indicators of compromise (IoCs) on macOS systems. It addresses the growing challenge faced by organizations as cyber threats become more sophisticated and stealthy, often leaving behind only faint traces—such as suspicious log entries, unusual network connections, or unexpected processes—that signal a breach. 

Key features: 

• AI-powered detection: AI processes and analyzes vast volumes of data at speeds faster than humans, enabling identification of patterns, anomaly detection, and adapting to new threats, providing an edge over traditional, manual security analysis.
• Modular agent architecture: built around a team of specialized AI sub-agents, each focused on a specific domain—system logs, network activity, or running processes—mirroring the structure of a real-world security operations center. The modular design enables easy extension: new sub-agents with specialized capabilities can be added as threats evolve or organizational needs change.
• Collaborative workflow: the sub-agents operate within a coordinated workflow, calling appropriate tools and sharing findings. This ensures a comprehensive and unified response to potential threats, increasing the speed and accuracy of detection.

Combining the speed and intelligence of AI with the expertise of human analysts, this agent-based approach represents a new paradigm in cyber defense—one that is adaptive, collaborative, and proactive—empowering organizations to detect, investigate, and respond to threats faster and more accurately than ever before. 

From experimentation to real-world solutions

The top projects exemplify how fast developers can build functional, real-world AI workflows by combining NeMo Agent toolkit’s orchestration, contextual memory, and tool integration features with open APIs, Omniverse’s industrial AI and data interoperability libraries, and domain-specific accelerators like cuOpt. 

Whether it’s orchestrating logistics agents that reason about forklift capacity or building productivity agents that review code or plan vacations, these projects show that you don’t need a massive team or budget to create.

If you’re curious about agent-based AI, now is the time to dive in. The NeMo Agent toolkit is open source, well-documented, and designed for experimentation. Try the examples, remix the agents, and see how far your ideas can go.