AI Makerspace - Guardrails AI Event

What are Guardrails?

Guardrails are specialized systems (typically models) that help "catch" when AI applications behave outside of desired parameters. They serve as safety mechanisms that monitor and validate AI outputs both pre- and post-generation to ensure applications behave as intended.

Think of Guardrails as quality control checkpoints that help protect against:

• Off-topic responses
• Data leaks and privacy violations
• Inappropriate content
• Factual inaccuracies and hallucinations
• Competitive mentions
• Jailbreaking and prompt injection attacks

About This Workshop

This hands-on workshop explores essential AI safety mechanisms through the Guardrails AI platform. You'll learn to implement and configure various types of Guards that are crucial for production AI applications.

What You'll Learn

Through practical examples in the guardrails_notebook.ipynb, you'll explore six critical categories of AI safety:

1. Topic Restriction - Keep AI conversations on designated topics
2. PII Redaction - Automatically detect and redact personally identifiable information
3. Content Moderation - Filter profanity and inappropriate language
4. Factuality Checks - Prevent hallucinations by validating responses against provided context
5. Competition Monitoring - Avoid inadvertent promotion of competitors
6. Jailbreaking Detection - Identify and block prompt injection attempts

Why This Matters

As AI systems become more prevalent in production environments, implementing robust safety measures is essential for:

• Compliance - Meeting data protection and content standards
• Brand Protection - Maintaining appropriate tone and avoiding competitor mentions
• User Trust - Ensuring accurate, relevant, and safe interactions
• Risk Mitigation - Preventing misuse through prompt injection attacks

Whether you're deploying customer service bots, content generation tools, or AI assistants, these Guardrails techniques are fundamental for responsible AI development.

Setup Instructions

1. Install Dependencies

uv sync

2. Configure Guardrails API

uv run guardrails config

Provide your Guardrails AI API key, found here.

3. Install Required Guards

uv run guardrails hub install hub://tryolabs/restricttotopic
uv run guardrails hub install hub://guardrails/detect_jailbreak
uv run guardrails hub install hub://guardrails/competitor_check
uv run guardrails hub install hub://arize-ai/llm_rag_evaluator
uv run guardrails hub install hub://guardrails/profanity_free
uv run guardrails hub install hub://guardrails/guardrails_pii

4. Get Started

Open the guardrails_notebook.ipynb file and follow along with the interactive examples. Each section demonstrates a different type of Guard with practical, real-world scenarios.

You'll also need an OpenAI API key for some of the Guards that use OpenAI models as backends.

Workshop Structure

The notebook is organized into focused sections, each building your understanding of AI safety mechanisms:

• Setup & Configuration - API keys and environment preparation
• Guard Demonstrations - Hands-on examples of each Guard type
• Real-world Scenarios - Practical applications and edge cases
• Integration Patterns - How to combine multiple Guards effectively

Key Takeaways

By completing this workshop, you'll understand how to:

• Implement multiple layers of AI safety validation
• Configure Guards for your specific use cases
• Handle Guard failures gracefully in your applications
• Build more trustworthy and compliant AI systems

Start your journey into responsible AI development by opening the notebook and exploring these essential safety mechanisms!