AI agents in production
An agent inside a production CMS translates pages, answers content questions and edits blocks. Startups get the same order: find the work a team repeats, build for it, then check that people still use it.
Dan Arkhipov
AI Engineer
I build AI applications and agents, from the interface to production.
A neuron hears thousands of signals and answers with one. An orchestra turns many lines into a single breath. A model reads everything and chooses the next word. I work somewhere between the three.
TypeScript · React · Next.js · Node.js · PostgreSQL · MCP
01 What I do
Full-stack applications
Design, interface, backend, deployment and logs, built end to end. Most things break at the seams between them, so it helps to hold them together.
Knowledge systems
Agents that keep research, notes and signals searchable. It began as a second brain for my own life, and the same idea can serve a business.
Testing new AI tools
New models, tools and agent skills get tried as they come out. What holds up goes into lectures. What breaks is usually more interesting.
02 Making now
Open source
Family Graph
A self-hosted archive for a family: a tree, a page for each relative, shared editing, and AI agents that help tend it through MCP.
For families who want to keep their own history.
In personal use
Writing system
Finds ideas for posts, drafts them and checks each draft. A person still has the last word.
For my own writing on LinkedIn.
In development
Mayaq
Turns signals from trusted public sources, such as social media and podcasts, into briefs about which ideas are growing.
For builders, marketers, analysts and writers.
03 About
I started on IBM mainframes, moved to interfaces, and now build full-stack applications and AI agents. One field kept leading to the next: the brain, data, human behavior, psychology. At some point they stopped being separate.
A model's attention and a listener's. Memory in files and memory in a head. An orchestra and a team of agents, each with its own line to hold.
I write music for orchestra and organ, and I sing. A score teaches that most of the work is deciding who stays silent, which turns out to matter with agents too. The rest of the time goes to reading, to people who know more than I do, and to whatever came out this week.