Peleke Sengstacke

Secure learning architectures for AI agent systems.

agent security adaptive retrieval exploit research observability Old Norse

I build secure learning architectures for AI agent systems so they safely improve over time. The learning side is qortex (adaptive retrieval) and buildlog (a measurable feedback loop that teaches coding agents to suck less). On the security side, I do threat models, sandbox architectures, and exploit research.

Currently: consulting on agent architecture and AI security for enterprise clients, and running a focused research program on adaptive retrieval and agent sandboxing. Looking for a founding engineering role where security and agentic workflows intersect.

Previously: Principal Engineer at 2U/edX, where I built the sandbox platform behind their $60MM+ ARR Cybersecurity Bootcamp. Before that, founding engineer at Trilogy Education: first technical hire, built their Cybersecurity Bootcamp from scratch (became their fourth product line), then transitioned to engineering leadership through the $750M acquisition. Technical writer at Scotch.io, Tuts+, DigitalOcean. I've been explaining hard things clearly for a decade. Now I build them, too.

Some of the writing on this site is co-authored with Claude. They handle code archaeology, structural drafts, and adversarial review. I handle narrative, voice, and editorial judgment. The author notes on each article say who did what.

I study Old Norse and Austronesian linguistics for fun. The sagas and joys of CHamoru are far richer than most of what this industry has to offer.

References on request.

"Difficile est saturam non scribere." — Juvenal, Satire I

(It is difficult not to write satire.)

← Home