I'm Garth Puckerin — an AI / Agentic Systems Engineer building agentic AI for the enterprise domain I know from the inside. I design and ship multi-service platforms: a code-comprehension graph and a governed memory layer running in production beneath my coding agents; a policy-bound decision engine that lets a model advise but never act; and an event-driven integration layer built to solve a real credit-union integration problem.

Underneath the AI work sits 18 years administering and integrating enterprise learning and workforce platforms — Docebo, Workday, SumTotal, and SAP SuccessFactors — across banks, pharma, and hospitals, plus a self-taught database habit going back to 1992 (FoxPro to Access to SQL/PostgreSQL to Neo4j) and a grants database I built for the grants management office at NYU's School of Education in 1998. That domain depth is the moat: I know exactly where enterprise learning systems break, and I build systems that account for it before they ship.

The platform is real and I keep its status honest. The comprehension graph and the memory layer are in production, self-hosted on a Synology DS1821+: my agents query a structural map of more than 100 of my repositories instead of reading whole files, and the decisions behind the code outlive the session that made them. The decision engine is released at v2.1 — one request in, exactly one directive out, nothing executed; the integration layer is a release candidate, built to solve the exact UKG, Axonify, LinkedIn Learning, and Docebo integration seams I hit on the job, though it was never deployed there. The capture pipeline — fragments from AI conversations netted into traceable project notebooks, every graph edge a citation — is deployed on that same hardware and in release stabilization. Their names drop one at a time, every Thursday through October 1, on the reveal wall.

My edge is the domain depth most AI engineers lack: I know what GxP and banking compliance require of software, and how to land working systems with real stakeholders from requirements through UAT to production. I develop daily with AI assistance — Claude Code, Codex, and Cursor — and still take on LMS administration, learning-technology, and integration contract work, where those two skill sets reinforce each other rather than compete.