I architect and build production software, developer systems, and computational platforms with AI as an implementation multiplier. The pattern is consistent: make the problem explicit, turn assumptions into testable contracts, use agents aggressively inside bounded scopes, and keep verification and promotion independent of the worker.
Recent work includes Ratchetloop, an auditable software-delivery pipeline for coding agents; Strategy Assayer, a large-scale computational research and evaluation platform; Asset Factory, a generative-AI production and publishing system; and Meetup Finder, a deployed product built with an architecture-first, multi-model delivery process.
That work sits on top of a long production-engineering track record across enterprise and public-sector systems, including Verizon, Kaiser Permanente, Walmart, and state government. Published research, five granted speech-recognition patents, industry recognition, and AWS ML / AI certifications are documented in Technical Credentials.
What I Build
Production Software
Web applications, APIs, workflow systems, integrations, automation, cloud services, and the supporting test and deployment machinery required to operate them.
AI-Assisted Engineering Systems
Coding-agent pipelines, deterministic validation gates, independent model review, local-model routing, run provenance, and the controls that turn fast generation into inspectable software.
Computational & Evaluation Platforms
High-throughput experiment engines, optimization, reproducible evaluation, provenance, and promotion systems where a plausible result is not enough and evidence has to survive review.
Architecture & Modernization
Turning manual processes, legacy systems, and loosely connected tools into explicit, maintainable software systems with clear interfaces, evidence, and operating boundaries.
Selected Work
Ratchetloop - Auditable AI Software Delivery
A local-first software-delivery pipeline for AI coding agents. A bounded task goes to a coder in an isolated Git worktree, deterministic checks validate the change, a different model family reviews the diff, and Ratchetloop leaves a branch plus an evidence-rich run record for a human merge decision.
The public implementation supports resumable runs, scoped commits, process cleanup, usage and velocity accounting, idea-to-plan-to-code execution, and validated Grok, Codex, and Claude providers. It intentionally does not merge, push, or deploy the target repository.
Strategy Assayer - Computational Research & Evaluation Platform
The largest system I have built independently: a research platform that separates search from proof. It combines a parallel experiment engine with tiered evaluation, walk-forward holdout, randomized controls, null distributions, multiple-trial correction, blocking promotion gates, and hash-stamped provenance.
A 679,140-point parameter space converged in 611 evaluations in under three minutes against an ~88-hour exhaustive-grid extrapolation. The private platform has produced 25,000+ experiment plans, 86,000 walk-forward windows, and 5.2 million evaluation runs. Systematic trading is the current workload; the engineering problem is general computational research and trustworthy evaluation at scale.
Deep technical case study · Public orchestration-engine slice
Asset Factory - Generative AI Production & Publishing
A local-first multimodal production platform that turns model inference into a controlled product workflow: persistent GPU model serving, image and audio generation, multi-signal quality scoring, evolutionary optimization, human curation, packaging, metadata generation, and marketplace publishing.
The operator dashboard spans Discover, Scores, Generate, Review, Publish, and a Custom Art Studio. By mid-April 2026 the project documentation recorded 23 themes, 991 theme entries, and 19 schema migrations. Assets Forge is one public-facing brand powered by the private platform.
Meetup Finder - AI-Native Product Delivery
A deployed mobile web product built as a compact demonstration of architecture-first, AI-assisted delivery: product and privacy boundaries first, independent model review of the design, agent-assisted implementation, automated tests, real-device validation, and deployment.
The first working implementation followed the initial design the same afternoon. The project is useful evidence of the other side of the portfolio: not infrastructure for its own sake, but a complete product moving from idea to a live beta.
Python Algorithm Interview Work Sample
Ten representative algorithm patterns implemented through Ratchetloop, with focused unit tests, deterministic reference-oracle validation, independent cross-model review, and GitHub Actions. It is intentionally a software-engineering work sample rather than a completion exercise.
github.com/t-espy/leetcode-python · See the delivery system behind it
View the broader project portfolio →
Selected Writing
A Discipline for Building Production Software with AI Agents
The engineering method behind the work: explicit architecture, durable decisions, verified external claims, bounded agents, deterministic gates, independent review, and evidence that survives the session that produced it.
Putting Qwen3.8-27B to Work on a DGX Spark
Engineering experiments on model routing, speculative decoding, harness overhead, output strategy, and what classes of real software work can move from frontier agents to local compute.
Autonomous Improvement Rate
A framework for reasoning about iterative autonomous systems in terms of throughput, information gained per iteration, and how effectively the system retains and exploits what it learns.
View writing and technical notes →
Production Engineering Track Record
- Enterprise and public-sector delivery: production systems for Verizon, Kaiser Permanente, Walmart, and state government environments.
- Published research: IEEE fuzzy-logic control paper; NASA conference and international workshop publications on genetic algorithms; U.S. Department of Energy technical report based on Duke M.S. thesis research.
- Granted intellectual property: co-inventor on five speech-recognition patents across U.S., European, and Chinese grants.
- Industry recognition: lead-development work on two systems recognized by Speech Technology Magazine as Most Innovative Solutions.
Supporting references are in Technical Credentials.
2026 engineering output: roughly 500,000 source-code insertions across 34 GitHub repositories, plus approximately 200,000 test and 130,000 documentation insertions. Generated output stats available on request.