SGInnovate x NVIDIA Workshop: How to build an AI Agent

Course Description & Learning Outcomes

Participants will benefit from two complementary learning experiences:

1. In-Person Hands-On Workshop (Registration Required)

Join a 3-hour workshop with NVIDIA experts’ guide, where you'll build and explore the first 02 modules of the learning path through guided hands-on exercises.

2. Continue Learning at Your Own Pace

After the workshop, participants will receive exclusive access to the complete learning path, including remaining 4 modules. Continue exploring advanced topics such as agent evaluation, customization, deep agents, and agent safety through self-paced learning on NVIDIA Brev.

What Will Be Covered in the In-Person Workshop

Module 1: Build an Agent

Learn the fundamentals of AI agents by building a Report Generation Agent from scratch.

What you'll build: An intelligent system that researches any topic, creates outlines, writes detailed sections, and compiles professional reports automatically.

Key concepts:

• The four core components of any AI agent (Model, Tools, Memory, Routing)

• ReAct architecture for tool-calling agents

• Building agents from scratch and with LangChain

• Using NVIDIA Nemotron models

Module 2: Agentic RAG

Evolve from basic RAG to intelligent agentic RAG systems.

What you'll build: An IT Help Desk agent that dynamically decides when and how to search knowledge bases to answer user queries.

Key concepts:

• Traditional RAG limitations and how agents solve them

• NVIDIA NeMo Retriever (embeddings and reranking)

• Vector databases with FAISS

• ReAct agents with retrieval tools

Continue Learning After the Workshop

Participants will receive access to the remaining modules in the learning path:

Module 3: Agent Evaluation

Master the art of measuring and improving agent performance.

What you'll learn: How to systematically evaluate agents using industry-standard metrics, LLM-as-a-judge techniques, and NVIDIA models.

Module 4: Agent Customization

Specialize agents for specific domains using synthetic data and reinforcement learning.

What you'll build: A bash agent customized into a LangGraph CLI expert using NVIDIA NeMo Data Designer for synthetic data generation and GRPO (Group Relative Policy Optimization) for training.

Module 5: Deep Agents

Build autonomous agents that handle complex, multi-step tasks with planning and delegation.

What you'll build: A production-grade deep agent with explicit planning, hierarchical sub-agent delegation, persistent memory, and sandboxed execution using Docker.

Module 6: Agent Safety

Build an OpenClaw personal assistant agent that executes inside and outside of an Openshell sandbox, complete with network and filesystem policies that demonstrate how the NVIDIA NemoClaw reference stack improves agent security.

By completing 06 modules, you'll be able to:

• Build agents that use tools, maintain context, and make intelligent decisions

• Implement RAG systems that dynamically retrieve and use information

• Evaluate agent quality using quantitative metrics and qualitative assessment

• Use NVIDIA technology including NIM, Nemotron models, and NeMo Retriever

• Customize agents through synthetic data generation and reinforcement learning

• Build deep agents with planning, delegation, and sandboxed execution

• Secure agents with kernel-level enforcement, data classification, and red-team evaluation

• Deploy and monitor agents in production environments

Continuously improve agent performance through systematic evaluation