HAKAN // DEMİRCİSelected work
Applications and open-source components developed to explore how AI agents make decisions, what permissions they act under and how they work with people.
A rule-based decision component for agent actions
Evaluates an action proposed by an agent and the supplied approval information against defined rules. Returns “allow”, “block” or “human approval required”.
v0.2.0 open-source component · Apache 2.0 license.
It does not execute actions itself or stop them in external systems. Authentication and enforcement of the decision are the responsibility of the system it connects to.
Hakan Demirci’s visitor agent
Developed to help visitors learn about Hakan’s work and clarify a business need through conversation. It uses approved information, asks questions to understand the scope and directs visitors who want to get in touch to a contact channel.
An application you can try on this site.
It does not send messages, create appointments or make price commitments on its own.
The long-term research and development ecosystem in which I work on authority boundaries, human approval, memory and auditable behavior in AI systems. Policy Engine is one of the public components of this work; Tales is its application that interacts with visitors. The ecosystem is under development.