Building a Responsible AI Tax Assistant Agent with Guardrails

In this practical workshop, you’ll build a GenAI-powered Tax Assistant Agent—capable of answering tax queries, retrieving documents, and handling tool calls. But here’s the twist: you’ll also learn to secure it using state-of-the-art responsible AI guardrails. From blocking unsafe queries to filtering PII and stopping jailbreaks, you’ll implement protections using tools like Nemo Guardrails, Llama Guard, Cleanlab, and Fiddler. Expect hands-on activities, simulated attacks, and real-time observability as you learn to build LLM agents that are not just smart—but safe.

Key Takeaways:

• Learn how to build a GenAI-powered tax assistant using LLMs and Retrieval-Augmented Generation (RAG).
• Apply AI safety tools like Nemo Guardrails, Llama Guard, and Cleanlab to secure agentic applications.
• Explore real-world threats like prompt injection, PII leaks, and unsafe queries—and how to defend against them.
• Gain practical skills in setting up guardrails, moderation flows, and observability for responsible AI deployment.

Anuj Saini is a Subject Matter Expert with over 15 years of industry experience in Natural Language Processing, Search Technologies, Statistics, Analytics, Modelling, Data Science, Data Mining, and Machine Learning. He has a proven track record in developing systems based on NLP, Machine Learning, and Cloud Technologies. You can reach him on LinkedIn.