I am a systems builder focused on human‑AI collaboration and AI architecture. My work centers on designing systems where language models operate within governed environments that preserve truth, continuity, and structured reasoning.
I have developed an interconnected portfolio of systems: AiMe (a persistent human‑AI partnership system), the COGS framework and its in‑house trained specialists, licensable governance infrastructure (SBA Spine, Runtime Governance Layer, Relational Integrity Coefficient), Ethos (value‑under‑pressure analysis), and applied products including Anvil and Corevah. My work explores the idea that intelligence in AI systems emerges not from models alone, but from the architecture surrounding them.
I enjoy building systems that turn complex technical capabilities into simple, reliable experiences for real users. My approach emphasizes truth as an architectural property, model‑agnostic infrastructure, and clear human‑centered interaction.
Architected and built a persistent AI system designed for long‑term human collaboration. The system separates model capabilities from system governance, allowing language models to operate within a structured architecture that preserves continuity, memory, and truthfulness. AiMe integrates persistent conversational memory, system‑level governance layers, and model‑agnostic infrastructure to support reliable human‑AI interaction over extended periods of work.
View AiMe →Designed COGS (Coordination of Governed Specialists), a framework that decomposes AI cognition into purpose‑built specialists: specialists gather information, a governance layer decides what is authorized, and the language model renders only what has been approved — no specialist governs its own output. Under this framework, individual cognitive roles (intent classification, value scoring, generation, judgement) have been moved off borrowed frontier models onto specialists trained in‑house, several of which now match or exceed frontier models on their benchmarks while running on local hardware.
View COGS →Built the Self‑Bounded Authority Spine, a deterministic governance layer that wraps a single language‑model call in two gates: an authority engine that decides what the system is permitted to say before the model speaks, and a compliance validator that drops any output exceeding what was authorized. Every decision is captured in a turn‑level provenance record for full auditability — a licensable component for AI deployments in regulated environments.
View architecture →Developed a runtime governance layer that injects a behavioral specification — valid outputs, boundary cases, abstention thresholds, and prohibitions — into every cognitive cycle, rather than relying on prompts or retraining. Demonstrated by lifting an intent‑classification task from roughly 20% to 100% accuracy with zero model changes and zero training data, improvable by editing a text file and re‑benchmarking.
View RGL →Designed the Relational Integrity Coefficient, a deterministic five‑subscale integrity score computed for every AI response from observable behavior rather than model self‑report. RIC measures groundedness, calibration, transparency, helpfulness, and pressure resistance, providing the kind of traceable behavioral measurement that emerging AI regulation requires.
View RIC →Designed and implemented a 7‑layer behavioral extraction pipeline that determines what a system actually values from its demonstrated behavior under pressure, not from what it claims. Ethos uses deterministic extraction layers, resistance scoring weighted by pressure intensity, and a P1/P0/APY value taxonomy, with architectural safeguards against hallucination and interpretive drift. 934 passing tests, published on SSRN, deployed live at trust‑forged.com.
View Ethos →Developed a framework focused on making truthfulness an architectural property of AI systems. Verum uses verifiable sources, structured retrieval, and system‑level provenance tracking to ensure responses can be traced back to evidence rather than model inference alone.
View Verum →Designed and implemented an intelligent infrastructure monitoring system that moves beyond traditional alert‑based monitoring. Instead of relying on static thresholds, Marshal analyzes contextual signals — trend velocity, temporal baselines, correlated anomalies — to detect developing incidents before thresholds are breached. The system operates through a deterministic Confidence Gate that evaluates signal significance, action confidence, and operational risk to determine one of four response tiers: Autonomous, Notify & Act, Recommend, or Observe. All actions are written to an append‑only audit log prior to execution, ensuring complete system accountability.
View Marshal →Built Anvil, an open‑source developer tool that brings discipline to AI‑assisted coding. A single Rust binary with a full terminal UI drives a real agentic coder that reads, writes, and edits files directly in a repository, with two human review gates — a plan gate and a phase gate — where two different model families critique the plan and the resulting diff. Cross‑provider by design (local Ollama through to the major cloud vendors) and released as public beta under Apache‑2.0.
View Anvil →Designed and shipped Corevah, an AI‑powered operations platform for field‑service contractors. Its assistant, Cora, handles scheduling, estimates, invoicing, and customer communication through natural conversation, with instant property‑based estimates, direct Stripe payouts, crew management via simple links, and an included professional website. Built from direct experience running a contracting business.
View Corevah →Founder and lead architect of an independent AI systems initiative focused on human‑AI collaboration and system architecture.
- Designed and implemented architecture for multi‑component AI systems
- Developed frameworks emphasizing truth, traceability, and system governance
- Built licensable governance infrastructure (SBA Spine, Runtime Governance Layer, RIC) for AI deployments in regulated environments
- Trained in‑house cognitive specialists that match or exceed frontier models on their benchmarks while running on local hardware
- Integrated multiple model providers into model‑agnostic infrastructure
- Designed interaction layers translating complex AI systems into usable human experiences
- Published research and technical writing exploring AI architecture and human‑AI partnership
Owner and operator of a residential and commercial property maintenance business. Responsibilities include client management, operational planning, logistics, equipment maintenance, quoting and estimating, and service execution. The business operates as a self‑managed enterprise requiring strong prioritization, reliability, and direct customer engagement.
Worked within a family‑operated residential care facility supporting daily operations and management. Responsibilities included facility maintenance, logistics coordination, and assisting with operational oversight in a care environment requiring reliability and attention to detail.
Before founding Canady Lawn Service, worked across a range of independent technical and skilled trades roles including construction and carpentry, horse training and farrier work, equipment maintenance and repair, and independent contract work across multiple trades. These roles developed practical problem‑solving ability, self‑direction, and the habit of learning new technical domains quickly.
Interest in computing began in the early 1980s with the Commodore VIC‑20, leading to a lifelong interest in computers, systems, and technology. This foundation eventually evolved into the independent development of modern hardware extensions for classic Commodore computers under the commodore4ever name — including Wi‑Fi modems and dual‑display communication devices — and ultimately into the independent development of modern AI systems beginning in November 2024.