AI Systems Architect · Independent Researcher
Deterministic control for
autonomous systems
I design deterministic control layers for autonomous and multi-agent systems — the layer that decides whether a machine’s decision is allowed to reach execution. Founder of ZVAKTHOR.
Profile
Twenty years of building software systems — the last two and a half dedicated to deterministic control for autonomous agents. Most agentic AI stops at a working demo. The hard part is everything after it: execution governance, observability, and failure handling. That is the part I work on.
My research asks where authority over an autonomous decision should live, how it is bounded and revoked, and how every decision — accepted or rejected — can be recorded and replayed. The component that can influence a decision should not be the component that authorises it.
Current work · ZVAKTHOR
An algorithmic execution platform for global financial markets. Specialist agents analyse markets independently; an isolated Rust core validates every decision for risk and execution eligibility before any order reaches the market. Every decision — accepted or rejected — is recorded and replayable.
Signal generation and candidate search.
Independent specialist agents propose decisions.
Isolated Rust core validates risk and eligibility. Deterministic stop.
Control boundary — nothing crosses unvalidated
Status
Core architecture and the first agent group validated end to end in a demo environment. Next phase: full agent activation and live-account validation.
Built on the same principle
Three systems, one principle: models propose or explain — deterministic engines decide.
Decision intelligence for AI architecture
Turns AI project planning into evidence-based, traceable decision packages covering architecture options, cost, risk, and mapping against governance frameworks. Several models are compared; the output is a documented decision, not a chat answer.
StatusDeployed; agent behaviour tested over several months. Not yet validated under production load or with real users.
Investment decision intelligence
Deterministic engines score risk, economic regime and allocation; AI explains the decision and never makes it. No broker connection, no custody, no execution. Every decision is recorded in an auditable ledger.
StatusDeployed; agent behaviour tested over several months. Not yet validated under production load or with real users.
Research · SSRN 2026
Open source
deterministic-control-referenceThe three-layer control boundary from Paper 03: revocable authority, an append-only decision journal, deterministic replay.
reliable-ai-workflow-engineRule-based autonomous workflow with an enforced lifecycle and an audit trail.
Contact
Open to conversations with investors and partners in AI infrastructure and quantitative finance.
hadi@hadisoufi.com