AI Systems

The fastest way to evaluate this skill is to use it: the chat on this site isn’t a widget I installed. It’s a retrieval system I built, running on a knowledge base I engineered, answering questions about me from structured, governed content. The medium is the credential.

Where this comes from

I build and operate the Strategic Intelligence Engine (SIE) — a system that takes a business’s knowledge (products, expertise, policies, voice) and makes it retrievable, chat-able, and publishable across a fleet of live WordPress sites. It runs today on my own e-commerce brands, answering customer questions against real inventory. Not a prototype, not a demo reel — production, with real buyers on the other end.

Building that taught me the two lessons that define how I work with AI:

Architecture beats prompts. When an AI system misbehaves, the instinct is to write better instructions. That’s usually wrong — and there’s a diagnostic tell: the instructions keep getting longer. Escalating prompts are the signature of a structural problem being managed with words. The durable fix is nearly always architecture: better-structured knowledge, clearer retrieval paths, tighter scopes. I wrote this up as the prompting fallacy, because it applies to far more than AI.

RAG is a knowledge problem, not a model problem. Everyone’s retrieval demo works; production retrieval is where they fail. Mine is structured-first — the system exhausts deterministic, structured lookups before vector search, which serves as backstop rather than backbone. The unglamorous work of schema design, semantic summaries, and synthetic questions is where retrieval quality actually comes from. That craft is knowledge engineering, and it’s the half of AI nobody puts in the keynote.

How I think about it

• Production or it didn’t happen. An AI system that hasn’t survived real users, real data drift, and real maintenance is a hypothesis.
• Integrity is architectural. My systems refuse to fabricate — “I don’t know” is a designed behavior — and no agent I build gets financial or destructive authority without a human in the loop. These are hard constraints, not tone suggestions.

The proof

The strongest proof is interactive — ask the system about me. The engineering story is in my knowledge base: the SIE, building AI agent systems, and knowledge engineering. And these are among the skills I employ full-time in my current in-house role, where AI-assisted development with production-grade guardrails is part of the day job.