Peleke Sengstacke

Agent infrastructure that learns safely.

I build AI agent systems that improve from usage without becoming liabilities.

Here you'll find adaptive retrieval that sharpens over time. Threat models and sandboxes that keep the learning loop from going sideways. Observability that traces every decision so the fixes actually stick.

Infrastructure

Adaptive retrieval, per-tool sandboxing, decision tracing. The stack behind the products and the security research.

Architecture + hypotheses →

Products

Tools only ship when I prove they work on real users. Here's where we test them before they hit the wild.