About
I'm Melanie Kertley. I run an AI team.
Not a metaphorical one — a real team of agents, each with its own context, its own job, its own handoffs. They help me build software, manage a small business, write a memoir, run a household, and do work that one person used to need a department for. I orchestrate them. They get on with it.
This is the part of my work I'm willing to show.
The varied resume
Before AI, I had a resume that didn't fit on a page. Cosmetologist. Au pair in France for two years. Nanny in seven different homes. Manager at Target. Restaurant management. Currently, administrative assistant at an accounting advisory firm — Action Advisors — handling client-facing work through tax season.
Pattern recognition was my hidden skill the whole time. Every industry runs on the same handful of problems, dressed differently. Different uniforms, same structural failures. Different bosses, same human dynamics.
It turns out that's exactly what AI orchestration needs — someone who can see the structure under the surface, name the job clearly, and hand it to the right tool.
How I found AI
I discovered Claude.ai in late 2025 and laughed out loud — that's my brother's name.
My brother Jon Claude — JC — has been an AI Orchestrator longer than I have. He's a madman white hat hacker in the best sense of the word: someone who lives in systems most people don't know exist and uses that knowledge to make things better instead of worse. When I started building, he was already there. Some of the architecture I use, he taught me. Some of it, we figured out together.
Family team. Real one.
How I actually work
Here's the part most "AI consultants" online don't talk about, because they don't actually do it:
I design agents. Each one gets a name, a clear job, and the context it needs to do that job well. Some live in my filesystem and remember everything across conversations. Some are summoned for specific tasks and forgotten when done. They have rules about what they touch and what they leave alone. They hand off to each other when the work moves between domains.
That's the work. Not "I use ChatGPT efficiently." Designing the roles, writing the system prompts, building the memory architecture, choosing the right model for the right job. Knowing when Claude is the right tool, when Gemini is, when ChatGPT is, when DeepSeek is, when the answer is none of them.
If you've ever managed people, this is more like that than like coding.
The receipts
SanctumTools
A production AI mental health platform I built using my own AI team. It exists because I have rapid-cycling bipolar I and the existing tools didn't meet me where I was. So I built one that did. It runs on Claude API + Firebase + the orchestration patterns I designed. Real users, real product.
Fred
The personal AI assistant my brother got me started on, that I shaped into the thing I needed: persistent file-based memory, daily mood tracking, file-by-file context, automated session-saving, mood and pain logging for SSDI documentation. Not a chatbot. A working colleague who remembers.
My stack
Everything runs on Google Workspace as the substrate.
The models — Claude (Anthropic, my primary build partner), Gemini (Google, chat layer + image generation + strategic review), ChatGPT (OpenAI), DeepSeek (long-context analysis), and others as fit demands — are all available to me on tap. I don't pick a side. The right tool depends on the job. Most of the work isn't in the model — it's in how I tell the model what to do.