KT Gilcrease
Technology & AI Consulting
KT Gilcrease

Twenty-four years turning complicated systems into software that actually works.

Technology Leader · AI Strategist · Systems Architect

I help organizations figure out what AI can actually do for them, and build the team, process, and guardrails to do it without losing control.

A working portfolio of systems, experiments, and things I was curious enough to build.
Chapter one

Nobody Assigned Me the Vision.

I didn't start in IT. I started part-time in accounting, doing invoicing, at a company called WIRB. Even then it was the largest IRB in the world, long before it became WCG, and it's just as much the largest one now.

A few weeks in, my manager handed me the manuals for our Solomon accounting system on a Friday and said he thought I might like to read through them over the weekend. Apparently I did. I came back Monday with a list of changes: workflow fixes, template redesigns, things that ended up shaving real, measurable time off our monthly close. Nobody asked me to fix the accounting system. I just couldn't leave it alone once I understood how it worked.

Before long, the IT director himself came down to ask if I'd build an interim time-tracking system in Access while they figured out a permanent replacement for the one they were phasing out. It was supposed to last a year. What I built tracked hours, ran weekly and monthly timesheets, and generated reports for every hourly employee in the building. It was still running five years later, long after I'd moved on to other things.

That other thing was a junior developer role in IT. They assigned me straight to a consulting team from Arthur Andersen, brought in to help build what would become IRIS. Weeks later, Enron happened, and Andersen dissolved around us mid-project. We kept a couple of their people on just long enough to get the project across the finish line. At its peak, 22 people worked on the team. I was the only woman in the codebase, and years after go-live, I was the last one still standing.

Along the way, I took a detour that turned out to matter more than I expected. I spent about six months in Bermuda, consulting for a bank and helping migrate their operations off a mainframe and onto .NET. Reconciliation logic, check-type microcoding, reporting, a global payment system juggling currencies, exchange rates, and fees that shifted by card network, location, and time of day. It was proof, to myself as much as anyone, that what I'd built wasn't specific to one industry or one company. The instinct traveled.

I came back and kept climbing: senior engineer, then manager, then director. As director, I created the architect role the org had never formally had, hired for it, and still reviewed and signed off on every piece of architecture that came through it. Over the following years, I reshaped how the core system scaled across acquisitions without erasing anyone's identity. I made sure it could be segregated hard enough to satisfy contracts as sensitive as Department of Defense work. And I built in enough flexibility that the organization could reorganize itself without ever needing to tear the system apart to do it.

Twenty-one years, one system, four titles, and not one of them was ever assigned to me before I'd already started doing the work.

24+ yearsSoftware & systems
From nothingHands-on builds
AI as collaboratorNot a shortcut
End to endConcept → architecture → build
Chapter two

Building It Myself Again

Twenty-one years of climbing a title ladder taught me how to design systems for other people to build. The last few years have been about proving I can still build the thing myself, from nothing, in languages I picked up on the fly, and about learning what it actually means to work with AI instead of just around it.

AI Planning Room
ROLES
BUDGETS
PHASE GATES

AI Planning Room

A local command center where I run Codex, Claude, and Gemini as a working team in one shared room. Each model gets a real job description and a definition of done for whatever role I assign (engineer, tester, architect), and I swap those roles when the work or the budget calls for it. Together we write the work orders, break them into phases, and stop at every gate to review and test before moving on. Nothing runs without my approval.

AI Orchestration · Node.js Read the full case study →
Yare
PLATFORM
TASTEBUD(S)
DRAFT

Yare & TasteBud(s)

Yare is a platform for AI companions that each specialize in one narrow thing, with shared memory, relationship modeling, and a behavioral template that keeps every companion in its lane. TasteBud(s) is the first product built on it: a curated roster of hyper-focused culinary AI companions. The same engine powered a completely different app (Draft, for writing) in under a day of setup.

Platform · Case Study Read the full case study →
Requirements Compliance
REQUIREMENTS
USE CASES
REVIEW

Requirements Compliance Architect

In Life Sciences, everyone audits everyone, because the stakes are too high not to. And yet the vendors validating your systems often profit from housing your data too, a conflict of interest nobody says out loud. The Requirements Compliance Architect (RCA) is one piece of a full validation workflow, focused entirely on requirements review, built deliberately not to integrate with the systems it reviews. It can't be bribed, can't be talked into approving something, can't be pressured to look past a gap.

AI Product · Case Study Read the full case study →
Francis
CONVERSATION
BEHAVIOR
KNOWLEDGE

Francis

Most AI chat is built to take over the conversation. Francis started from the opposite question: can an AI be useful without needing control? It's a space to think out loud, not a therapist and not a generic chatbot, built on a philosophy I call continual return to self. The hardest part wasn't writing more instructions, it was learning that better constraints beat more of them, every time.

AI Product · Case Study Read the full case study →

How the work fits together

These projects were not built at random. Each one tested a different piece of a larger goal, and wherever I could, I tested the risky ideas where a mistake was cheap.

  1. 1

    Requirements Compliance Architect

    The first focused reviewer, built for requirements in regulated work. It stays independent of the systems it reviews and never approves anything on its own.

  2. 2

    Yare and TasteBud(s)

    Let me test a platform of focused AI assistants where a mistake costs a bad dinner, not someone’s safety. It taught me what belongs in a shared platform and what belongs in each product.

  3. 3

    Francis

    Taught me how to shape conversation and keep an AI inside clear guardrails, and how to test and version its behavior before it goes live.

  4. 4

    AI Planning Room

    Taught me to run several AI models as a team, with defined roles, cost limits, an audit trail, and a person approving every step that matters.

Together they point to a larger goal, AI work that stays accountable at any scale. A small team can keep it light. A regulated team can require every review, test, and sign-off. Either way, it is built around intended use and human accountability.

Work with me

Put AI to work without losing control of it.

A career spent inside regulated, audited systems taught me that the hard part is never just the technology. It's the process, the people, and the guardrails around it. That's where I come in.

Tell me what you're trying to achieve, and I'll tell you honestly whether AI is the right tool for it, and what it would take to get there.

Start a conversation LinkedIn See the projects ↑