Software engineer with 10+ years building reliable systems end to end across regulated, high-stakes domains: U.S. Air Force programs (held a Secret clearance), healthcare and insurance, fintech, and fleet payments. Beginning August 2025, on my own initiative, I committed to AI engineering in depth, well beyond day-to-day use: how large language models actually work, and how to make them reliable through operating directives and geometric representation. I architect and ship real systems with measured results, full-stack from database to cloud deploy, and I prove my work with tests and reproducible numbers.
On my own initiative and outside of employment, I committed to AI engineering in depth. I did not learn it on a job; I sought out not only how to use large language models but how they work internally and how to optimize them. I get reliable output from stock models by directing them with explicit operating directives that demand proof and reject unverified claims, the same discipline that took a verifier's hallucinated output from about 39% to 0%. I architected and built real, tested systems with this approach (below), and I developed an original framework, dimensional programming, built on z = x · y as an organizing manifold and exploring Fibonacci-based dimensional dynamics, with each idea demonstrated in a working tool and each claim labeled by how well it is supported.
Wrapped generation in a deterministic verifier that rejects unverifiable candidates, taking the rate of unsupported output reaching the user to zero while best-of-N kept the answer rate high. Measured, reproducible.
Deterministic, content-addressed context (bfx-ingest) maximizes prompt-cache hits, cutting repeated input cost in a typical iterative session. Proven by a reproducible benchmark.
Shipped a water-system telemetry and compliance platform with a signed go-live gate and a tamper-evident audit trail, covered by 123 deterministic automated tests. Live at theconduit.me.
Re-architected multiplayer netcode to event sourcing: state replays byte-identical from one seed, eliminating an entire class of state-divergence bugs. Fast Track verifies 503 paths with 0 teleports.
Python, JavaScript / TypeScript, C# / .NET, Java, SQL, GLSL
LLM integration, RAG, evaluation and verification harnesses, structured / JSON outputs, tool and function calling, model routing, prompt caching, token budgeting, hallucination reduction; self-hosted local models (Llama, Qwen, DeepSeek, Mistral, Phi-3) via Ollama and LM Studio
Node.js, Express, Spring Boot, REST APIs, microservices, WebSockets, PostgreSQL, SQL Server, Oracle, MySQL, SQLite, Linux, nginx, Docker, PM2, zero-downtime deploys
Azure DevOps pipelines, GitHub Actions, Azure, automated testing, observability
Angular, React, WebGL2, Three.js, HTML / CSS
Event sourcing, content-addressed storage, deterministic replay, HL7 / FHIR interoperability, secure SDLC in cleared environments
Apr 2024 to Present
Feb 2022 to Jan 2024 · Remote
Apr 2021 to Dec 2021
Personal projects built end to end to demonstrate the work. Each one runs, is tested, and carries a real number or a live link.
Dimensional programming / manifold-as-data: representing data as derivable geometry so localized context is computed on demand instead of stored or re-sent, the same intuition as the manifold hypothesis, embeddings, and autoencoders. Demonstrated, not only theorized: a dependency-free API measures a roughly 99.7% token reduction answering localized questions over a large nested structure. Published openly, with each claim labeled by how well it is supported, at dimensionalprogramming.com.
B.S., Computer Science and Software Engineering
Seeking an AI engineering or backend / platform role where reliability and cost-efficiency matter. Reach me through the contact page.