I build AI systems that run themselves in production — not demos.
Autonomous agents, LLM data pipelines, and automations that do real work — designed, built, and shipped solo, then hardened in production. A decade of building software; the last two years deep in agentic systems.
What I build
Autonomous agents & multi-agent workflows
Systems that run unattended — with a human in the loop at the decisions that actually matter.
LLM data pipelines
Messy, multi-source, real-world inputs turned into structured, scored, actionable output.
Automation & integration
n8n, APIs, cloud, Postgres, Kubernetes — the plumbing that makes it run every day, not just once.
AI that ships
Strategy that ends in working software, not a slide deck.
Selected work
AI delivery into an existing team Client
Embedded with an existing team and applied targeted AI/ML to a problem they'd assumed needed slow, expensive manual expertise: identifying and attributing collectible items by matching imagery against catalog and auction references. Shipped an end-to-end working pipeline — tangible results at a fraction of the expected cost and turnaround.
Computer vision · ML matching · end-to-end delivery into a live team
Multi-agent operations platform
A coordination layer for running many AI agents alongside a human across projects: shared task and knowledge stores, typed agent roles, cross-agent messaging, live dashboards, and scoped tool access so agents act through real, permissioned tools rather than free-form prompts. Agents pick up work across sessions, operate from shared context instead of starting cold, and escalate to a human at defined gates.
FastAPI · Postgres · MCP · Kubernetes
Autonomous cross-border scanner
A production pipeline that monitors messy, multi-language public data sources across several countries every day — rendering JavaScript-heavy pages, extracting structured fields from unstructured listings with LLMs, scoring each against a configurable thesis, and surfacing only the high-value few to a human via a unified live dashboard. Runs unattended; turns hundreds of noisy listings into a short ranked shortlist.
n8n · headless-browser rendering · LLM extraction · Postgres · scoring · alerting
Also: production trading infrastructure (risk gate, multi-venue execution, position accounting) · real-time stereoscopic view-synthesis from dual camera streams (FFT sync, optical flow, reverse-warp blending; WebGPU + server pipeline). When the problem is hard and algorithmic, not just an integration.
About
I'm a software engineer with a decade of building — startups through enterprise (Capital One) — focused for the last two years on production AI: agents and automations that run unattended and do real work. I work solo and remote-global, async across timezones, and I'm senior enough to be the technical lead in the room when a project needs one. I build things that ship and keep running.