About

I'm a full stack engineer based in Colorado who specializes in AI integration. I build software that puts large language models to work inside real products and real teams, and I write about what actually holds up once it ships.

Before AI was most of my job, I spent years building enterprise applications, which is where my habits around security, reliability, and accessibility come from.

AI integration, developer tooling, and automation for teams that cannot afford to get it wrong. The through-line is shipping AI features that are useful, safe, and worth what they cost to run.

The model is a component, not the decision maker. Anything that has to be right every time (permissions, approvals, money, data retention) lives in plain deterministic code, and the model works inside those rails. In my document governance MCP server, for example, Claude only explains a decision after a code-level approval gate has already made it.

I also pick the shape of an automation on purpose: a loop when the model genuinely needs to decide what comes next, a graph when my code should. Picking by accident is how a $6 a month job becomes a $400 a month one.

I review every line an AI writes, even when the tests pass. I encode security rules (OWASP checks, accessibility checks) as skills and hooks so the tooling enforces them instead of relying on anyone remembering, and I built vibe-check, a linter that catches the sloppy patterns AI code tends to ship with.

For AI features themselves: audit logs, human approval where the stakes call for it, and least-privilege access for anything an agent can touch.

Value per token, not token volume. Agent architectures can quietly burn hundreds of dollars a day, so I design for the cheapest model and the fewest turns that do the job well, and I measure before I optimize.

I started in .NET and C# on large enterprise systems and now work just as comfortably in TypeScript, React, Python, and Node. On the AI side: the Claude and OpenAI APIs, Model Context Protocol (MCP) servers, agents, RAG, and Claude Code tooling. On the platform side: Azure, Docker, Kubernetes, Terraform, and CI/CD.

I apply models rather than train them. My focus is integrating AI into production systems, not foundational ML research.

A working MCP server for document-retention governance, vibe-check (a linter for AI-generated code), a trigger-indexed memory layer for coding agents, and several small AI apps people actually use, including a recipe-to-grocery-list PWA built on Claude Vision. Most of it is open source on GitHub.

Here on hereshecodes.com, where I write short, practical pieces on AI engineering. I also post on LinkedIn (daniellescantling) and publish code on GitHub (hereshecodes).