Work with me

I build AI products from idea to paying customers, research how AI safety systems actually behave versus how they're documented, and figure out where prompt engineering ends and prompt architecture begins. If you're working on something in that space and want someone who's already made the expensive mistakes, I'm available.

What I do

AI Red-Teaming & Safety Evaluation

Adversarial testing of AI systems — finding where your content moderation breaks, where your safety guardrails have gaps, and where the model does something different from what your documentation promises. My methodology comes from offensive security testing applied to AI content systems, and I've published the research to prove it works.

Prompt Architecture & Engineering

SkeinScribe runs on a custom prompt architecture that handles branching narratives, entity tracking, content rating tiers, and multi-layer moderation — all on top of the Claude API. That's not a weekend project with a system prompt. If you need prompt systems that work reliably at scale rather than demo-well-once, that's what I build.

AI Content Moderation Design

Building or evaluating content safety systems, rating frameworks, and moderation pipelines. I've done this from both sides — building the moderation systems and testing where they fail. That combination turns out to be useful.

AI Product Consulting

I've taken an AI product from "what if" to live, paying customers on the Claude API. If you're building something similar, I can probably save you the time I spent learning what doesn't work. This includes architecture decisions, API integration, content pipelines, cost optimization, and the hundred small things that don't show up in the documentation.

How I work

Remote and async, based in Ottawa (ET). I'm direct, I don't pad timelines, and I'd rather tell you something won't work than let you find out the expensive way. I'm available for:

  • Advisory / fractional — ongoing, lighter touch. Good for teams that need an experienced perspective on AI product decisions without a full-time hire.
  • Project-based — scoped deliverable with a clear start and end. Prompt architecture builds, moderation system design, that kind of thing.
  • Red-team engagements — time-boxed adversarial testing of your AI systems. You get a report of what I found and how to fix it.
Let's talk →