Architected the core automation engine, end to end
From CRUD to a graph-driven execution engine in ~8 months, canvas to backend.
In context → Read(profile.yaml)⎿ identity · headline · summary
Read(profile.yaml)⎿ identity · availability · services · engagement
// Full-stack engineer · 20+ years · Charlotte, NC
Kevin G. Henderson
I build AI agents a business can trust — and I deleted my own when Claude did the job better.
Need AI agents, an MCP integration or a 0→1 build shipped — properly, end to end?
I'm Kevin — a full-stack engineer in Charlotte, NC, working remotely. For 20+ years I've taken products from blank repo to production. Lately that means AI agent platforms: the builders, evals, review flows and infrastructure that make LLM features trustworthy enough to run a business on. I adopted agentic coding early, in mid-2025, and it's how I work now.
stack TypeScript · React · Goalso Vue · Node · Postgres · TemporalRemote (US) · Charlotte, NCIC / Founding · no management
What is he great at?
Grep("pillars", profile.yaml)⎿ 4 matches, in priority order
01
AI product engineering in production
Shipped LLM agents with tool calling, evals that gate changes, human review, cost accuracy and prompt-injection hardening.
02
0→1 architecture, end-to-end ownership
Pick the stack, write the spec, ship frontend and backend, own it in production.
03
Staff-level multiplier
Design docs, technical evaluations, team conventions and agent tooling, ~1,000 code reviews.
04
AI-native velocity
1,700+ merged PRs in two years, and agent-built side projects far outside my core stack.
Show me the numbers.
Read(profile.yaml#stats)⎿ 4 outcomes · each traces to the experience page
- Lines retired when I deleted my own AI feature
- −28.5k
- PRs to ship Human Review, spec to production in five weeks
- ~70
- Engineers building on ZenGRC Pro, which I architected from the first commit
- ~10
- Merged PRs (Nov 2024 – Oct 2026)
- 1,705
What has he shipped recently?
Read(experience[0].highlights)⎿ Scratchpad / Clearskies · Full-stack engineer, Clearskies agent & workflow platform · Nov 2024 – now
An LLM agent that writes workflow graphs — then the call to delete it
Built it, made it reliable, then retired it for a plugin inside Claude and ChatGPT.
Read the story →Eval-gated prompt and tool changes
Structural scorers and an LLM user-simulator gate every prompt and tool change.
Read the story →Pausing autonomous agents for a human, durably
Agent actions wait on Temporal for a named reviewer — spec to production in five weeks.
Read the story →- ZenGRC · 2022–2024 · Principal architect, ZenGRC Pro · promoted Senior UX Engineer → Staff → Senior Staff
- Moovweb → Layer0 → Edgio · 2015–2022 · Six years across the platform, developer docs, open-source frameworks and customer console as Moovweb became Layer0 and then Edgio.
- Earlier · 2005–2015 · Virtual Instruments; Bank of America; Aym Technologies; CompanionCabinet Software; Solers, Inc.. B.S. Computer Engineering, Virginia Tech.
How does he work with agents?
Read(how-i-work)⎿ the loop · guardrails · where agents fail · the numbers
1,705 merged PRs in two years, most of them written by agents I directed — and the guardrails that make that pace safe: repo conventions, verification skills, eval gates, and build checks that refuse anything the evidence doesn't support.
Anything outside the day job?
Read(projects)⎿ 4 projects · all built agent-first, directed by Kevin
web · ios · androidScorecardReal-time multiplayer scorekeeper for 11 card games on web, iOS and Android — built and shipped solo, agent-first with Claude Code (Nuxt 4, Postgres/Drizzle, WebSockets, Capacitor, RevenueCat), with 180 test suites in CI.
wear osUnofficial Pixel FaceWear OS watch face plus phone/watch companion apps that mirror live notification icons onto the watch — Kotlin and Watch Face Format, a stack new to me, built entirely agent-first. Private repo — walkthrough on request.
macosMailTraymacOS menu-bar app unifying Gmail, Outlook, Yahoo and iCloud webmail with unread badges and notifications — Electron, Vue 3, SQLite/Drizzle, shipped as a notarized DMG and Mac App Store build. Private repo — walkthrough on request.
astrokevhender.netThis site: static Astro, generated from one typed profile file, with a grounded "Ask about Kevin" assistant on a Cloudflare Worker.
Where can I read the details?
Glob("case-studies/*")⎿ 4 case studies
01I built an AI workflow builder — then deleted itFrom an in-app chat builder to a headless MCP plugin inside Claude and ChatGPT; evals, native tool calling, and knowing when the best UI is someone else's.02Measuring LLM features before trusting themDeterministic scorers before LLM judges, a simulated user for clarifying questions, grading effects instead of claims, and shadow-testing an LLM's replacement on live traffic.03Pausing autonomous agents for a human, durablyHow to pause an agent before a consequential action and wait days for a human, without losing state.04ZenGRC Pro: from first commit to a ten-engineer productStack choices, a typed GraphQL layer, the Pages-vs-App-Router call, and a redesign with no design team.
How do I reach him?
send_message(to: "Kevin G. Henderson")⎿ paused for human review
Approval required — this action needs a human (you)
- 1.Yes, email kevhender@gmail.com
- 2.Message on LinkedIn — in/kevhender
- 3.Browse GitHub — @kevhender
- 4.Hiring? Paste a job description — get a grounded fit report
Press 1–4 to pick, Enter to confirm. A nod to Human Review: agents propose, people approve.
What could he build or fix for us?
Read(profile.yaml#services)⎿ 7 services · 4 featured · each mapped to shipped proof
Short engagements, part-time or fractional, remote. Scoped, fixed-outcome work preferred. I don't take work that competes with my employer's products.
AI agents & workflow automation
Design and ship LLM agents with tool calling, workflow engines, and human-in-the-loop review that a business can trust.
Proof: Clearskies Workflows · AI Workflow Builder · Human Review
MCP servers & AI integrations
Expose your product to Claude, ChatGPT and Gemini over MCP — scoped tokens, LLM-friendly error contracts, prompt-injection hardening.
Proof: MCP security · Frame AI drafts
LLM evals & reliability
Stand up eval harnesses that gate prompt and model changes, plus observability and cost accounting for LLM features.
Proof: Agent evals · Multi-provider LLM layer · Failure alerting
0→1 product builds
Go from idea to a production MVP — stack choice, architecture, frontend and backend, CI, deploy.
Also: Vue / React / Next.js rescue & upgrades · AI-native engineering enablement · Fractional Staff engineer — details
How do engagements work?
Read(profile.yaml#engagement)⎿ formats · process · rates
Formats
- 01Fixed-scope project
- 02Part-time / fractional retainer
- 03Short advisory or architecture review
Process
- 01
Intro call
- 02
Written scope & estimate
- 03
Build with weekly demos
- 04
Handoff with docs
rates Not published — let's talk.still open to Staff / Principal / Founding Engineer roles
Has he shipped this stuff?
Read(profile.yaml#stats)⎿ 4 outcomes · each traces to the experience page
- Lines retired when I deleted my own AI feature
- −28.5k
- PRs to ship Human Review, spec to production in five weeks
- ~70
- Engineers building on ZenGRC Pro, which I architected from the first commit
- ~10
- Merged PRs (Nov 2024 – Oct 2026)
- 1,705
Where can I read the details?
Glob("case-studies/*")⎿ 4 case studies
01I built an AI workflow builder — then deleted itFrom an in-app chat builder to a headless MCP plugin inside Claude and ChatGPT; evals, native tool calling, and knowing when the best UI is someone else's.02Measuring LLM features before trusting themDeterministic scorers before LLM judges, a simulated user for clarifying questions, grading effects instead of claims, and shadow-testing an LLM's replacement on live traffic.03Pausing autonomous agents for a human, durablyHow to pause an agent before a consequential action and wait days for a human, without losing state.04ZenGRC Pro: from first commit to a ten-engineer productStack choices, a typed GraphQL layer, the Pages-vs-App-Router call, and a redesign with no design team.
How do we start?
send_message(to: "Kevin G. Henderson", re: "project")⎿ paused for human review
Approval required — this action needs a human (you)
- 1.Tell me about your project — email kevhender@gmail.com
- 2.Read how engagements work — /work-with-me
- 3.Message on LinkedIn — in/kevhender
- 4.Browse GitHub — @kevhender
Press 1–4 to pick, Enter to confirm. A nod to Human Review: agents propose, people approve.
The “Ask about Kevin” assistant needs JavaScript — everything it knows is on this page, or email Kevin.