Denis nonened.io RU get in touch

AI agents that do the work. Not in a demo, in production.

I'm Denis, an engineer. Eight-plus years of building systems; the last few spent on LLM agents, the memory behind them and the infrastructure they run on around the clock. Below is a map of what is running for me right now.

Message me on Telegram or hello@nonened.io
A map of my AI infrastructure Telegram leads to two agents, Hermes and OpenClaw. Both use one shared memory, Hindsight, and one model gateway, LiteLLM, which routes requests to cloud models Claude, GPT, Gemini and Grok and to local models on Ollama. Coding agents Claude Code and Codex share the same memory and receive tasks from the agents. All of it, together with Postgres, DNS, monitoring and backups, runs on three servers in one Docker Swarm of 39 services. Telegramyou message an agent Hermespersonal agent no. 1 OpenClawpersonal agent no. 2 · 20+ channels Shared memoryHindsight · one for all Model gatewayLiteLLM · routing, limits, cache Cloud modelsClaude · GPT · Gemini · Grok Local modelsOllama · on my own hardware Coding agentsClaude Code · Codex ServicesPostgres · DNS · monitoring · backups 3 servers · Netherlands and Russia · Docker Swarm · 39 services · one GitHub Actions pipeline A map of my AI infrastructure Telegram leads to two agents, Hermes and OpenClaw. Both use the shared memory Hindsight and the model gateway LiteLLM, which routes requests to cloud and local models. Coding agents Claude Code and Codex share the same memory. Everything runs on three servers in one Docker Swarm of 39 services. Telegramyou message an agent Hermespersonal agent no. 1 OpenClawagent no. 2 · 20+ channels Shared memoryHindsight · one for every agent Model gatewayLiteLLM · routing, limits, cache CloudClaude · GPT · Gemini LocalOllama · own hardware Coding agentsClaude Code · Codex · same memory ServicesPostgres · DNS · monitoring · backups 3 servers · Docker Swarm · 39 services
This is not a "what it could look like" diagram. It is a live system: two agents with one shared memory, a model gateway, coding agents and almost forty services on three servers. I build the same kind of thing for other people's workflows.

Things you can open and try.

  1. Home page of the AI Engineering Roadmap course

    AI Engineering Roadmap

    An open course that takes a Python developer to building agentic systems. Five phases, 150+ materials, quizzes and progress kept right in the browser, no sign-up. Written in Russian; updated to the state of September 2026.

    Astro · MDX · PWA · in Russian

    roadmap.nonened.io
  2. The Kupi app on a phone: a shopping list sorted into categories

    Kupi

    A shopping list that sorts items into categories by itself. Dictate or type; the app understands typos and word forms and works offline.

    React · TypeScript · PWA · in Russian

    app.nonened.io
  3. Elsevia: a catalogue of interactive stories with a poster and an Explore this story button

    Elsevia

    An interactive romance game: night-time Paris, full-screen scenes, choices that change the story and a live conversation with the character powered by a language model. Authors get a mobile Studio. Open demo: five chapters, several endings, no sign-up.

    Expo · React Native · Node · PostgreSQL · LLM chat

    elsevia.nonened.io

  4. CRM for a consultancy in the UAE

    The working system of a company that sets up businesses and arranges visas in the Emirates. Clients, companies and shareholdings, document checklists, questionnaires, PDFs from templates, a knowledge base and one inbox for every channel.

    Django · Supabase · Docker · staff sign-in

    crm.nonened.io

  5. Accounting for a construction company of 80 people

    Timesheets with automatic file-format detection, overtime, payroll, payouts, per-site costs, printable forms and a complete change log. An internal system, so no link. But this is the kind of system I build.

    Django · React · PostgreSQL

    no public access

Four ways to work with me.

Agents in business workflows

Support, documents, requests, CRM, reporting. An LLM embedded in a real process, with memory, tools and clear limits on what the agent may do. Not a chatbot for show, but a system that takes routine off people's hands.

RAG · tool use · multi-agent setups · memory

Business and accounting systems

CRM, timesheets, payroll, documents, a single inbox. Classic development with AI built in where it pays off, not everywhere.

Django · PostgreSQL · React · Docker

Process review and rollout plan

I look at the process and say where AI will save time and money and where it is not needed. You get a plan with estimated timelines, costs and effect.

a low-risk way to start

MVP in 2–4 weeks

A working prototype of an AI product: interface, model, deployment. Something you can show to customers and investors before the big spend.

idea → a product you can touch

If AI will not help with your task, I will say so. That is a result too.

First we check that AI solves the problem. Then we roll it out for real.

Review

A call to walk through the process. I say honestly whether AI will help and what it would cost.

Pilot on your data

A prototype on real documents and conversations. You see the quality before investing.

Rollout

Integration with CRM, messengers, email and knowledge bases. Access, and training for the people.

Support

I keep an eye on quality and spend, and extend the system to neighbouring processes.

About me, in facts.

Eight-plus years of engineering: high-load systems, infrastructure, and in recent years neural networks and agents in production. I work alone, with no intermediaries: you talk directly to the person who understands the task and does the work.

The easiest way to see what I do is to look at what runs at my place.

  • Three servers in the Netherlands and Russia, 39 services in one Docker Swarm, deployed by a single GitHub Actions pipeline.
  • Two personal Telegram agents, Hermes and OpenClaw, with one shared memory: tell one, and the other knows.
  • Claude Code and Codex are wired to the same memory per project, so context survives between sessions.
  • My own model gateway: Claude, GPT, Gemini, Grok and local models, with routing, limits and a cache.
  • My own Postgres, my own encrypted DNS, monitoring, backups.
  • A personal health platform: lab results, workouts, genetics, reviewed by a model.
  • An open AI Engineering Roadmap course that I run and keep current.

AI Engineering Roadmap

I don't only build, I teach. An open course from Python developer to engineer of agentic systems. Free, no sign-up, progress stays in your browser. The course is in Russian.

phases
5
materials
150+
edition
04 · 09.2026
Open the course

Shall we talk about your task?

The first call is free. Send a couple of lines about the process you want to improve, and I will tell you whether it is worth putting AI into it.

Message me on Telegram or hello@nonened.io