engineer · agent systems
let's turn a boring workflow into an ai agent system.
tool-using agents with real boundaries, memory, and monitoring — built to survive contact with real users.
about
i build tool-using agents and the systems that keep them alive in production: typed tool contracts, permission boundaries, persistent memory, failure detection, and monitoring. the hard part is never the demo — it's everything after it.
i've been a founding engineer at an rl-environments company and i build my own agent systems independently. before ai i shipped web and smart-contract software.
skills
languages
agent systems
reliability
backend
selected work
labels distinguish operated systems, private source, and public code. links lead to the artifact that's actually available.
claudima
operated system · private sourcea self-hosted rust telegram system with json-schema-constrained host actions, bounded web-research subagents, persistent memory, and a human-reviewed browser bridge. the public case study maps its security claim to concrete boundaries and residual risks.
agent guard suite
public code · pythonfour small, zero-dependency python libraries i built for the boring parts of running agents: repeated-call protection, deterministic output checks, silent-failure signals, and token-budget accounting. each repository exposes a narrow api instead of pretending to be a full observability platform.
activity
writing
work with me
i'm open to co-founding, advising, and hands-on client work — building agent systems that have to run in production.
embed not loading? open the booking page ↗
contact
email is the fastest route.