I build AI-agent infrastructure and reinforcement-learning systems — and automate my own warehouse with them.
AI/ML engineer, self-taught. Last December I taught myself Python to stop my team counting shipping tickets by hand; that first tool worked. Within months I was shipping a stack of MCP servers on the official registry, a foundation RL model for workforce scheduling, and automation whose measured time savings add up to roughly 20 hours a week across warehouse logistics. I ship end-to-end and I'm honest about what my tools can't do.
A domain expert who taught himself to build and ship AI.
I run operations at a warehouse. Last December I got tired of watching my team count every shipping ticket by hand, twice, once for the pickers and once for the packers, when the numbers were already sitting online. So I taught myself Python to gather them automatically. That became MasterMetrics, a proof of concept that worked. From there I kept going, nights and weekends and a lot of it from my phone: reinforcement-learning agents that learn to schedule a warehouse workforce in simulation, a stack of Model Context Protocol servers now live on the official MCP registry, and more tools that automate my own job.
The thread through all of it: I understand the domain I'm optimizing, I build the whole thing (model → server → deploy → docs), and I don't oversell. That combination is what I want to bring to a team.
What I'm after: a full-time role in applied AI, forward-deployed engineering, or ML engineering — somewhere domain understanding and shipping speed both count. If that's a fit, let's talk.
What I build
Two flagships — and the work around them.
Flagship · AI-agent infrastructure
Xylem — a knowledge & coordination layer for AI agents
Run more than one AI coding agent on the same repo and two things break: they forget between sessions, and they overwrite each other. Xylem fixes both — with no server of its own, because git push already enforces "only save if nobody changed this first." A phone and a desktop become equal peers.
1,000simulated concurrent races: zero lost, zero double-granted
Records the why behind decisions in editable JSON, so a fresh session starts oriented. ~92% less context loaded vs. dumping the store; says "I don't know" instead of guessing.
Clark — a foundation RL model for workforce scheduling
A variable-shape transformer + LSTM (~18M params) trained with a 3-stage curriculum. Pretrain once, fine-tune a new facility in ~3.3 h. The hard part isn't scheduling one warehouse — it's one model that holds up as the workforce scales. That's the generalization result, from held-out simulation.
~93%order completion from 5 to 50 workers (sim)
Jack — PPO + LSTM with temporal memory: 98.2% order completion across ~9.4 simulated years on a single facility. The validated predecessor that Clark generalizes.
Balatron — An autonomous Balatro agent: PPO policy + a forward-simulating build planner + tactical heuristics, on a held-out fixed-seed harness. Confirmed white-stake wins — an active project, still training today.
Conductor — Local automation platform with a web dashboard — Redis workers, scheduled tasks, 24 MCP tools. Shipped as a Windows product.
waveform-MCP — 150+ tools giving Claude full control of a DAW — compose, mix, master, render. My largest tool surface.
llm-evals — Eval framework for LLM tools and agents — swappable executors + scorers, regression detection built in.
skillmatch-mcp — A job-fit analyzer that scores your resume + GitHub against a JD — yes, built during my own job search.
rag-pipeline — Fully-local document Q&A (llama.cpp + Llama 3.2 + nomic-embed). No cloud, no external vector DB.
clark-mcp — A fully-local natural-language interface to the Clark agent — MCP server + offline hermes3 client. No cloud, no API cost.
barcode-label-maker — Generates Avery barcode PDFs from an Excel pick-list, auto-shrinking SKUs, width-fit Code 39. Ships a synthetic catalog so real data stays private.
shipexec-metrics-tracker — Automates a manual daily shipping-metrics reporting workflow across the team with Selenium, pandas, and Excel — measured time savings of ~20 labor-hours a week. A proof of concept, validated against real ShipExec data.
From counting tickets by hand to distributed systems and foundation models.
Not a bootcamp. A full-time job, a phone, and every spare hour, starting last December. Each month I took on a genuinely harder domain. Tags are color-coded by thread.
Dec 2025 · the start
Taught myself Python to stop my team counting shipping tickets by hand. Built MasterMetrics, automating our daily operations metrics (a working proof of concept).
PythonSeleniumworkflow automation
Feb 2026
Went public — open-sourced my warehouse tools and wrote git-onboard to master Git.
Git internalsCLI designopen source
Mar 2026
Deep reinforcement learning — Dolly → Jack warehouse agents, plus Balatron (a Balatro agent, still training today).
Every number on this page, next to the thing that produces it.
Rendered from data/claims.json, which is the single
place any of these is written down. Check whichever one you doubt.
Not on this list: adoption. The registry listings are
published, not popular — my user count is in single digits, there's no production traffic, and
no code review from anyone but me.
Looking for applied-AI, forward-deployed, or ML engineering work.
If you want someone who understands the domain, builds the whole thing, and ships fast — I'd love to talk. Remote, US-based, and available now.