Datadog LLM Observability with Gemini

Protecting LLM powered applications requires visibility into how models behave, respond, and interact with data in real time. This hands on workshop focuses on using Datadog LLM Observability with Gemini to monitor and defend against modern AI driven security risks, including prompt abuse, unsafe outputs, and unintended data exposure. Participants will work with a pre instrumented application on Vertex AI to see how prompts, responses, evaluations, and traces reveal security and safety issues as they happen.

The session shows how correlating LLM traces with logs, metrics, and infrastructure context supports auditability, incident investigation, and risk reduction. Attendees will learn how to detect prompt injection attempts, monitor safety and toxicity signals, prevent sensitive data leakage, and validate that AI workflows remain grounded in approved data. The workshop is designed for teams responsible for securing AI systems and maintaining trust, control, and accountability in production LLM environments.