Practical Development, AI, and Complex Live Projects

I work with systems that can no longer be evaluated solely by their technology stack. Most of the interest begins where simple solutions stop working: an old project cannot be rewritten from scratch, AI needs to be embedded into a real business process, migration affects years of accumulated data, and a beautiful idea turns out to be significantly more complex than it seemed upon first analysis.

Currently, the bulk of my practice lies at the intersection of classical development, live web system architecture, and the practical application of AI.

What is Happening Now

One of the current commercial focuses is HappyBaby2000. This is an old online store that cannot simply be stopped and rewritten: it has a MODX backend, an accumulated catalog, SEO history, and working workflows. Therefore, the new public part is being developed on top of the old system, and technical decisions have to be made taking real legacy into account.

Work there right now is not about abstract "digital transformation," but about concrete things: price variations, quick catalog editing, redirects, sitemap, microdata, mobile interface, and the gradual development of an AI agent.

This project is a good illustration of one of my main specializations: figuring out a live complex system, preserving what already works, and gradually fixing what prevents it from evolving.

AI Without Magic

I work a lot with AI, but I do not train my own models. I am interested in the practical application of existing models and infrastructure: ChatGPT, OpenRouter, external APIs, local models, custom agents, memory, tools, and integration with working systems.

One of my current experiments is Lira: an AI agent with persistent memory and its own working context. This is no longer just a chat, but an attempt to build a digital employee who knows the projects, tasks, history of decisions, can work via API, and is gradually learning to participate more effectively in real work.

Practice quickly shows the boundaries of such ideas. It is not enough to give an agent access to data: you have to teach it to distinguish a fact from a hypothesis, understand the intent of the conversation, criticize reasonably, avoid exposing the internal structure of the system as useful client content, and turn discussions into actions.

The broader context of this work is related to haih-agent and ne-chatgpt.ru.

When "Rewrite from Scratch" is Bad Advice

A significant part of my experience is associated with legacy systems that continue to bring value and money many years later. In such projects, a new architecture does not start with a blank slate.

One must take into account old URLs, search traffic, business data, integrations, user habits, and dozens of decisions for which no one remembers why they were made.

This line extends from shopModx and shopModxBox through MODX → prisma-cms → freecode → haih-agent and today manifests in real migrations and support for live client systems.

Sometimes the Best Consultation Result is to Reject the Original Idea

Extensive experience is useful not only when you need to suggest how to implement something. Often, the initial framing already contains incorrect expectations: from AI, from rewriting the system, from automation, or from the cost of changes.

In such cases, my task is rather to quickly find the weak point in the assumptions, explain the real limitations, and help formulate a task that actually makes sense to solve.

I am more interested in practical results than in confirming a beautiful idea.

What Else is Developing

Aside from client work, I constantly test my own ideas on real projects. Active topics right now include the architecture of haih-agent, persistent AI agents, multilingual websites, old content migrations, stable human-readable URIs, and experiments with knowledge systems.

For example, recent work on human-readable URIs quickly went far beyond generating a nice slug: I had to account for old incoming links, percent-encoded addresses, 301 redirects, and internal linking of a live knowledge graph.

It is stories like these that gradually become the main content of this website: not a technology catalog and not a showcase of services, but real tasks, discovered problems, solutions, and conclusions that can help you understand how I work and where my experience can be useful.