NARVI
An applied-AI company — agent orchestration, AI safety engineering, and agentic security.
Founded and directed by Akeim Joseph. A.R.C.H.O.N., PRISM, and Anubis are documented here as they're built, not after the fact.
Applied AI Systems
Orchestration engines that make foundation models behave reliably: closed tool manifests, self-critique gating, deterministic evaluation harnesses that grade against fixed cases instead of vibes.
A.R.C.H.O.N.AI Safety Engineering
Anti-hallucination enforced by architecture, not prompted. A system that can't invent a capability it doesn't have, can't fabricate a coordinate it didn't compute, and refuses its own proposed actions under a bounded self-critique budget.
A.R.C.H.O.N. / AXFAgentic Security
A deterministic compliance gate for AI agents — judging the real consequence of a proposed action against hard, auditable invariants, before it executes, with no session history required to catch what matters.
AnubisHuman-AI Interaction
How a directed AI system should actually feel to the person using it — permission gates as real UI, not fine print; state that's honestly shown, not simulated.
PRISMThis is where the engineering behind NARVI's systems gets written up directly — architecture decisions, what broke, what got cut, and the research discipline underneath all of it. Documented as it happens, not after the fact.
Latest
Human-AI Interaction: Transparency Is the Load-Bearing Trust Variable
Recent workplace-AI research identifies agent transparency as the single strongest predictor of user preference — ahead of accuracy, speed, or capability. Most agentic UIs still treat it as an afterthought.