LLM-powered apps and agents flows behave like a new entry point into your system: similar to an API, but far less predictable.

The best place to manage that risk is before a request ever reaches the LLM, using an “incoming guardrails” layer.

This implementation-minded talk starts with a quick refresher on how LLMs work, what “agents” are, and what people mean by agentic flows.

We will then shift into a practical, software-engineer-friendly approach for adding incoming guardrails before requests hit a model.

We will cover common checks like prompt-injection, malicious intent, toxicity, and out-of-scope requests, as well as how to recognize higher-risk cases like potential self-harm or medical emergencies and route them to an escalation path.

You’ll leave this talk with a simple reference architecture you can adapt to your own stack.

Speaker

Eyal Wirsansky is a Staff AI Engineer at Aingelz Inc., and an AI adjunct professor at Jacksonville University.

Agenda

6:00-6:20: Networking

6:20-6:30: Introductions

6:30-7:30: Main presentation

Online Link

This will be a hybrid meetup. You can join us online at:

https://meetn.com/eyalscommunity