Hire Bret Kerr · AI GTM Engineer
GTM tooling built with Claude Code: skills, agents and pipelines where agents propose and deterministic gates dispose.
Ships production code with Claude Code, not just prompts: skills, subagents, verification gates.
Has built GTM and job-search tooling end to end: ApplyForge, Dashboard Builder, MMPipe.
Instruments pipelines so numbers are recomputed, not asserted: Transformer Atlas, Skill Memory Bank.
The checklist · typical requirements, each mapped to a live proof
Seven Claude Code skills for the MoEA deep-research pipeline: paste messy research, get an interactive dashboard plus three publish-ready assets in about 60 seconds. Apache 2.0.
Open Accelerator Hackathon, Lane 2 Best in Show. Agent Build Day 2026, June 27, Fort Point, Boston. Red Hat · IBM · Mass AI Hub.
A local-first iOS job-search co-pilot: track every application in a Cockpit, batch-tailor résumés and cover letters in the Forge. Free, bring your own key. See also the Dashboard Builder at /labs/job-prep-dashboard-builder.
A typed rewrite system that drives three skills into an auditable research tree, a labeling harness (/MoEA-labeler), and a captured Sonnet reasoning trace (/SonnetSkillReasoning).
GPT-2 small through TransformerLens: form a circuit hypothesis, ablate a head, rerun the forward pass, measure. Every number is recomputed from the real residual stream.
About ten hours of interview tape compressed into four source-anchored explainers, six pairwise analyses, and one reusable URL-to-explainer skill.
Three deep proofs
Hackathon · Best in Show
A governed memory control plane that compiles conversations into SKILL.md packages.
Proves · Can scope, build and demo an agent system in a day, with a judging panel as the gate.
Lane 2 Best in Show · Red Hat / IBM Open Accelerator Hackathon, June 27 2026
Open it live →
Pipeline · video in, instrument out
Gemini ingests the YouTube URL natively; Claude builds a source-anchored explainer; a correlation matrix lets the interviews argue with each other.
Proves · Multi-model orchestration shipped as a reusable skill, not a one-off demo.
~10 hr of tape · 4 explainers · 6 pairwise analyses · 1 skill
Open it live →
Mechanistic interpretability · runnable
GPT-2 small in the browser: form a hypothesis, ablate an attention head, and measure whether the prediction moves.
Proves · Comfort with model internals and with instruments whose numbers are recomputed, not quoted.
Every number recomputed from the real residual stream
Open it live →How I work · Signal → Thesis → Test → Ship
Opposed deep-research briefs (Gemini as cartographer, GPT as adversary), then cite-checked and reconciled.
One claim per piece, anchored to primary sources and written in the house voice. Story Extraction and Content Architecture, from the manifesto.
Deterministic gates, not model consensus, decide what ships. Quotes are verified, numbers are recomputed, and goldens are locked.
One operator. Cloudflare Workers, a parity-gated deploy, and a public commit history.
The full methodology: Agentic Content Marketing →
Contact
Email is fastest. Everything linked from this page is live, so you can click into any of it before we talk.