i am jonathan haas.

i'm interested in making AI systems more reliable through better evaluation. currently building behavioral testing infrastructure for language models—adversarial testing, robustness under distribution shift, and converting measurements into guardrails and feedback signals.

security engineering taught me how fragile production systems can be. i worked as an engineer at Snap, Carta, and DoorDash, then built ThreatKey to help companies catch issues before they exploded. that experience drives my current work at EvalOps, my applied research lab: turning model evaluation into a first-class primitive so we can ship reliable, steerable AI.