Governance that proves itself under pressure.

Felix Montanez

Co-Founder · Runtime Architecture

Most AI systems fail in production for a simple reason: they accept outputs without verifying them. I have been focused on a different approach, building AI systems where outputs are treated as claims, and must be validated before they are allowed to become real.

Over the past year I designed and validated InnerLight OS, a verification-constrained execution runtime that enforces this at the system level. The core invariant is simple:

• Only states that are admissible at time t are allowed to become real
• Inadmissible states are non-executable, non-influential, and non-realisable

This is not enforced through prompts or behaviour. It is enforced through topology: a non-bypassable execution boundary, a runtime-owned commit authority, no state mutation outside the verification gate, no path where invalid states can influence future outcomes.

I have implemented this as a working multi-agent runtime that executes full software life-cycles, from architecture to implementation to QA to verification, with enforced structure and deterministic outcomes.

My background spans high-consequence engineering, distributed systems, full-stack development, and production startup environments. I served in the U.S. Navy as a nuclear-trained operator working with submarine reactor systems, where verification, procedure, and execution discipline were not abstract concepts. I later carried that systems mindset into startup engineering, including work connected to will.i.am's technology ventures, where speed only matters if the system can actually hold together.

As a Navy veteran and member of the Hispanic community, I bring a practical engineering view to OBEXGATE: governance cannot live only in policy documents or advisory layers. It has to survive contact with runtime, teams, pressure, and real deployment.

Amelie Kingsbury Barry, PhD EE, PhD CS

Co-Founder · Managing Director, Niti Logic

I work across systems that most people encounter only from one side.

I watched my mother lose to systems she did not understand before I had words for what I was seeing. She immigrated legally from Thailand speaking almost no English, brilliant and resourceful, and completely unprepared for what it meant to exist inside bureaucracy. By the time I was in school I was the one making phone calls, translating not just language but logic. I could not always get her through. But I paid attention.

The question I have spent my career answering is the same one I could not answer for her as a child: what does this system actually require, and what do you do next?

Nearly thirty years later, that question runs through all of it. Boolean retrieval and early NLP at Pubnetics and RealLegal. Seventeen years on satellite power systems. The administrative and financial architecture of U.S. healthcare across payer, provider, and IPA layers. Permaculture studies that sharpened my understanding of interdependent systems, constraints, resilience, and unintended consequences. Two PhDs, in Electrical Engineering and Computer Science.

As a female and mixed-race founder, I have spent much of my life moving between systems: cultural systems, technical systems, healthcare systems, bureaucratic systems, and ecological systems. The work has always been the same at the deepest level: understand the wider context, identify the real constraints, and architect solutions that can function for humans inside the system, not just on paper.

The common thread is not the domain. It is how rule-based systems behave when decisions must translate into real-world outcomes. If a decision is not traceable, transparent, and defensible at the point of impact, governance has failed.

I solve for that boundary.

OBEXGATE is the product expression: a deployed runtime governance enforcement product, built because advisory governance does not survive contact with deployment. Niti Logic is the parallel deployment, applying the same architecture (PRISM) to healthcare decision logic.

I do not build advisory models. I build systems that can be used.