Chief Executive Officer & Chief Architect
Tony Richards has spent 38 years building systems where the cost of failure is measured in money, safety, or lives — and where “probably correct” has never been an acceptable standard.
That career began in 1987 with the US Army’s Enlisted Distribution and Assignment System: an expert system that matched hundreds of thousands of soldiers to duty stations based on skills, career progression, and unit requirements. It was human-in-the-loop workforce allocation AI — built on formal logic, explicit rules, and verifiable decisions — years before those terms entered the mainstream technology conversation. The experience established a conviction that has shaped everything since: that systems with real consequences require deterministic foundations, not probabilistic ones.
From there, Tony moved through every major infrastructure inflection point of the past three decades. He designed early VoIP registration servers and network protocol stacks at a time when real-time UDP handling was an unsolved engineering problem. He co-designed a foreign exchange trading engine at eSpeed capable of processing over one million orders per second — earning a co-authorship on US Patent 2008/0172318 for Aggregated Order Books — where the difference between a working and broken lock-free data structure was measured in microseconds and the difference between a correct and incorrect trade was measured in millions of dollars. He architected industrial IoT gateways at Honeywell, building the fault detection, isolation, and recovery systems that keep building infrastructure running when networks fail, hardware degrades, and the physical world refuses to cooperate with software assumptions.
The pattern across all of it is the same: every domain that truly cannot fail solves reliability through formal constraints, explicit rules, and systems that can prove their behavior rather than merely describe it.
Tony founded Nandeshou on the conviction that the AI industry is building mission-critical systems on the wrong foundation. The current generation of AI tools — including the agentic systems now being deployed at enterprise scale — are built on probabilistic engines that cannot formally verify their own outputs, cannot prove why they made a decision, and cannot guarantee that their behavior tomorrow will match their behavior today. For consumer applications, this is acceptable. For the systems that run businesses, manage finances, and control physical infrastructure, it is not.
The Mushin architecture is his answer: a neuro-symbolic framework that formally separates creative reasoning from deterministic execution, uses Goal Contracts to translate intuitive plans into verifiable specifications, and produces software that can be audited, proven, and trusted. Not because the AI is well-prompted, but because the architecture enforces correctness at the structural level.
Tony is a builder-CEO in the most literal sense. The Nandeshou platform is built with the Nandeshou platform. Kanshin handles identity. AethOS manages the infrastructure. Takumi synthesizes the code. Mushin verifies it. A five-person team is shipping what larger organizations cannot, because the tools compound: each capability built with Nandeshou becomes part of the substrate that builds the next one.
His writing on deterministic AI, the limits of probabilistic guardrails, and the architecture of verifiable intelligence appears in the Nandeshou Insights series.
