LLM features in production
Features on Claude, OpenAI and Gemini with structured outputs and function calling. Evals, guardrails and human review go in before launch.
// Nate Arnold · full stack + AI engineer
15+ years shipping production software. Today I build LLM features that hold up in production, and I ship code alongside coding agents that plan, build, test and review in parallel.
// how I build with AI
Features on Claude, OpenAI and Gemini with structured outputs and function calling. Evals, guardrails and human review go in before launch.
Several coding agents working in parallel, each in its own git worktree, wired to issues, code and monitoring through MCP. I review every diff.
// side project
🌗 Every family has its phases.
Co-parenting schedules, shared expenses and reminders for a blended household. One household can hold several co-parenting arrangements, each with its own kids, custody pattern, expense split and approval threshold.
Household partners see everything. Co-parents see only their own arrangement, enforced in the database with Postgres row-level security rather than in app code.
// what I'm up to
Full stack features for an AI-powered construction site documentation platform, in TypeScript across React, Next.js, React Native and Node.js.
Full stack apps, LLM features, agentic workflows and RAG assistants that connect models to client data, from scoping through launch and support.
// before that
Member app and coaching platform in React Native, React and GraphQL on AWS, including LLM coaching features.
Co-ran a small consultancy building websites, Jamstack apps and AI integrations for clients in healthcare, education, ecommerce and SaaS.
An electronic medical records platform for provider practices, in TypeScript, React and GraphQL.
Rental and fleet management for construction equipment, a shared design system, and React Native for field crews.
iPad apps for pharma sales teams, with native iOS plugins in Objective-C.
Responsive front ends for major brands at a software consultancy.
Nine years teaching high school science. It still shapes how I mentor and review code.
// stack