Building Secure and Trustworthy LLMs Using NVIDIA Guardrails

Course details

Guardrails are essential components of large language models (LLMs) that can help to safeguard against misuse, define conversational standards, and enhance public trust in AI technologies. In this course, instructor Nayan Saxena explores the importance of ethical AI deployment to understand how NVIDIA NeMo Guardrails enforces LLM safety and integrity. Learn how to construct conversational guidelines using Colang, leverage advanced functionalities to craft dynamic LLM interactions, augment LLM capabilities with custom actions, and elevate response quality and contextual accuracy with retrieval-augmented generation (RAG). By witnessing guardrails in action and analyzing real-world case studies, you'll also acquire skills and best practices for implementing secure, user-centric AI systems. This course is ideal for AI practitioners, developers, and ethical technology advocates seeking to advance their knowledge in LLM safety, ethics, and application design for responsible AI.

Concepts

Introduction

Leverage guardrails to build secure LLMs
What you should know
Exercise files and setting up your environment

Foundations of Guardrails for Language Models

Introduction to guardrails for LLMs
Behind the scenes - How Guardrails enforces LLM safety
Understanding the role of embeddings in safeguarding LLMs

Crafting Conversational Guardrails

Defining conversational boundaries
Advanced Colang flows
Guardrails in action

Integrating Custom Actions

Enhancing LLMs with custom actions
Building a custom inquiry action

Retrieval Augmented Generation (RAG) with Guardrails

Retrieval-augmented generation (RAG) with actions
Securing LLMs against sensitive topics
Debugging and optimizing Guardrails

Real-World Applications and Best Practices

Case studies - Guardrails in action
Best practices for implementing guardrails
Future of guardrails and LLM safety

Conclusion

Next steps in building LLM applications