I build AI systems that solve specific business problems — end-to-end and solo. The consulting version of my last company taught me that I'd rather be building systems than running sales motions. I'm looking for AI implementation, applied AI, and solutions engineering roles at companies with defined problems and real users — places where I can build full-time.
Systems I've Built
Local-deploy Laravel app that serves HTML compliance training courses and enforces completion integrity through an append-only SHA-256 audit chain, verifiable without external dependencies. Optional Bitcoin anchoring via OpenTimestamps for clients requiring third-party verifiability. Small by design — no SaaS lock-in, no vendor trust required.
A set of composable custom Claude skills that convert raw SOP documents into narrative training modules — from document analysis through script generation, instructional design, and visual production. Modular architecture allows the pipeline to be updated stage-by-stage without rebuilding the whole system.
AI video segmentation tool built on Meta's SAM 3 — full product with Stripe payments, video upload pipeline, mask generation, and serverless GPU deployment on RunPod. Natural language prompting for rotoscoping. Shelved after recognizing Adobe would ship a native click-to-track equivalent bundled free in their creative suite. Validated the technical direction; learned that UX integration into existing professional workflows matters more than novel interaction patterns.
Reference implementation for Claude Code + MCP: an outbound prospecting agent wiring Apollo (prospecting), Firecrawl (company research), and SQLite (pipeline state) into a coordinated workflow through custom Claude skills. Built to test whether a solo operator could run a credible outbound motion through Claude Code alone. The interesting technical result is how cleanly MCP lets a single skill orchestrate four heterogeneous tools without glue code. I use it now as a reference when thinking about multi-tool agent architectures.
AI Tools & Architecture
A custom Claude skill that forces structured reflection across an entire conversation before producing output. Catches model biases, surfaces missed considerations, and produces more intellectually rigorous responses than standard prompting. Applicable to any high-stakes Claude deployment where output quality matters more than speed.
Domain-agnostic learning loop built as custom Claude skills — assess, practice, introduce, reassess. The same pedagogical architecture powers implementations across unrelated subjects, proving the pattern generalizes. Each skill specializes Claude for a specific domain while sharing the core instructional engine.
Training & Courses
Three-module interactive lockout/tagout safety training course with comprehension checks, built from a 24-page OSHA procedure in one weekend. Proof-of-concept for the SOPs production pipeline.
Six-module course on human-AI interaction fundamentals — why language fluency triggers us to treat models as entities, how to calibrate expectations, and how to write your first context document. Grounded in original research on the cognitive science of human-AI perception. Learner outcome: a working context document that establishes their relationship with the model. Backward design — the course exists to produce that artifact, not to lecture.
Background
8,500+ live English teaching sessions — three years in Korea, then online to learners worldwide. The practical foundation for everything I now know about how people learn, resist, and eventually adopt new systems — which turns out to be most of what makes AI implementation succeed.
MA in Creative Writing. BA in Communication. Comfortable talking to a plant manager about lockout/tagout procedures and to an engineering team about retrieval architectures in the same afternoon.