Guardrails & Evals: How to Build Hard Constraints for AI
And crash-test them until you can trust they hold
Explore Live Slide Deck on Vercel
Full interactive keynote slides, live framework diagrams, and code snippets.
01 // Session Abstract & Core Thesis
In an era of non-deterministic LLMs, conventional unit tests fail to catch probabilistic regressions. This talk introduces an engineering blueprint for establishing hard boundary constraints on LLMs, generating automated red-teaming datasets, and running continuous evaluation suites in CI/CD pipelines.
02 // Key Takeaways & Actionable Frameworks
Why assertions fail on generative output and how semantic evaluation replaces them.
Building multi-layered guardrails: input filtering, schema adherence, and output validation.
Crash-testing LLM agents against prompt injection and model drift before reaching users.
Available for conferences, technical workshops, and panel keynotes.