The demo was never the hard part
Everyone is adding AI right now, and a demo takes an afternoon. The hard part is the 20% that decides whether anyone still uses it once the novelty wears off: retrieval that stays current, evaluations that catch regressions before your users do, and automation that fails safe instead of silently doing the wrong thing. I've spent fifteen years shipping production systems, and I hold AI to the same standard.
I help teams in two ways.
AI-accelerated delivery
A small team that leans hard on AI throughout the development loop can move like a much bigger one. I'm doing precisely this today — building a full creator platform from scratch (subscriptions, courses, live streaming, billing) with a handful of people and an AI-first workflow, shipping in a fraction of the usual time. If you want to move faster without ballooning headcount, this is the model.
Production AI & automation
I build the AI features and automations that run in production: LLM-powered workflows, retrieval over your own data (RAG), evaluation harnesses so changes are measured rather than guessed at, and agentic automation that takes real work off people's plates. I've shipped AI features inside a live consumer product on AWS Bedrock, and wired LLMs into systems where both correctness and cost matter.
I build the tooling AI agents run on
Under Arjia Labs I maintain open-source tools like clu, a local-first tracker for coordinating AI coding agents, and yori, a library for managing AI prompts, agents, and skills. That's the layer beneath the chatbot — and it's the layer that breaks when you try to put agents into production.
How I work
I'll tell you where AI helps and where it's the wrong tool, scope a concrete first deliverable, and ship something measurable instead of a slide deck. Boring, reliable foundations; AI where it earns its place. If you want to ship faster with AI, or get an automation into production without the hype, let's talk.