Software development · AI adoption · Consulting · Fractional CTO
More than twenty years of commercial software development — from national-scale information systems and platforms with intricate financial logic to hardware of my own design.
20+ years
shipping commercial software
10 industries
from oil and healthcare to social networks
30+ projects
taken to production
I build systems from scratch — and sort out the ones already running: what to do about the architecture, where AI belongs, who owns the engineering side.
Development
01
Systems the whole business runs on: storefront and customer accounts, an admin panel with roles and permissions, heavy catalogues with search, warehouse and logistics, loyalty and promo, reporting for management. Server-side rendering where speed and search visibility matter.
02
Exchange with government and corporate information systems, courier services, payment gateways across several countries, digital signatures and electronic invoicing. Guaranteed delivery, outbox, idempotency, observability.
03
iOS and Android sharing a codebase with the web: push notifications, offline mode, code scanning, live updates that skip the store review queue.
04
Reverse engineering closed protocols, custom boards and firmware, radio links, connecting industrial equipment to a mobile app. From the logic analyzer to a production run.
05
Making sense of an inherited system and planning a staged migration: a dependency map, tests around the critical paths, the move itself without stopping the business. Working through a large codebase and covering it with tests is done with AI agents — that saves weeks. .NET Framework → .NET 10, AngularJS → modern Angular.
06
The schema is designed for the growth ahead, not for the volume you have today. Transactional load is kept apart from analytical: PostgreSQL for operations, ClickHouse for heavy slices and reporting. On top of that, visibility into what is happening right now and agent access to the data over MCP — a question in plain language, an answer out of the warehouse.
Artificial intelligence
07
Embedding language models and agents into your system so they earn their keep instead of staying a demo: an assistant over your own data, document and correspondence processing, automation of routine operations. MCP servers wired to internal databases and services, streaming responses, control over request cost and answer quality.
08
Working out where AI genuinely pays off and where it turns into an expensive toy. An audit of how the work is done today, a shortlist of tasks with measurable return, a pilot on your real data, and a staged rollout plan. Document and correspondence handling, customer support, reporting — wherever it earns its keep.
09
Moving a team onto agents: project rules and architecture written down beside the code, tooling wired to the database, the browser and the test suite, project commands for recurring work. Review stays a hard gate — what grows is speed, not the number of defects in production.
10
A licence plate and vehicle recognition model of my own — trained for the job rather than taken off the shelf. Data collection and labelling, training, tuning for real shooting conditions, and running on a live video stream. Detection and classification wherever packaged solutions fail on your footage.
Consulting & leadership
11
One-off expertise where a wrong call is expensive: choosing the architecture and stack for the job, an honest assessment of the system you inherited, an audit of how your vendor actually works, technical due diligence before an investment. I answer straight — including when the right answer is "do not build this".
12
A part-time chief technology officer for a company that needs CTO-level expertise but not yet a full-time workload. Architecture and technical decisions, hiring and growing developers, release planning, talking to investors and vendors in their own language. From a few days a month.
Commerce & events
One system the whole business lives in: sales and warehouse, courier delivery, promo and loyalty, iOS and Android apps. A payment and settlement stack of its own — acceptance through more than ten providers across several countries, internal balances, reward calculation and payouts, digital signature and invoice filing into the state system. A ticketing subsystem of its own: venue seat maps, seat booking, ticket generation and validation at the door. The server-rendered storefront and the role-based admin panel sit on top of all that.
Gov integration
Ingesting data from weighing stations, validating it and guaranteeing delivery into government information systems. Separately, the equipment itself is watched: an agent on the station reports the state of cameras, load cells, induction loops and the WIM controller, so wear and degradation show up early rather than when the station goes down. A web panel accounts for every message and every station.
Hardware
There is no documentation for the radio module, so the protocol was recovered from the signal itself. Then a custom board for the stock slot and its firmware: weight reaches the tablet with no extra hardware and no modification to the scale.
Computer vision
A model of my own instead of a packaged one: data collection and labelling, training, tuning for real conditions — night, rain, dirty plates, speed. Runs on a live video stream alongside road enforcement systems.
Other work
A national information system for the Ministry of Health of Kazakhstan, an accounting system for a defence agency, risk management for the oil industry, a business consulting platform with an AI assistant and expert review of its answers, an application of my own for launching and hosting AI agent sessions, an international education platform, a training course delivered through a mobile app, a social platform for content creators at a European startup, an integration gateway, a real estate portal, an auto parts marketplace, a warehouse management system, a knowledge testing platform, school management — and others.
Below are some of the tools that have been in use on recent projects.
AI & computer vision
Backend
Frontend
Mobile
Data
Infrastructure
Hardware
This is how I work myself, every day. Agents sort the mail, open tasks, move documents along their routes and watch that nothing stalls. The same thing gets built into a company — into your systems, not instead of them.
An incoming message is read for meaning: what is being asked, by whom, what is required and by when. A task appears in the tracker with a description and an owner, and the right people are notified. A person steps in where judgement is needed, not where text has to be retyped from a letter into a form.
Invoices, acts and contracts are read, checked for completeness and passed along their route — into the accounting system, to the bookkeeper, into the archive, to the counterparty. Discrepancies surface on the day they arrive, not at month-end reconciliation.
An agent keeps an eye on what is going on: which letters went unanswered, which tasks have not moved, where the process stalled and who is already being held up by it. The summary arrives on its own, without being asked for.
Agents work where the data already lives: your mail, your tracker, your databases. Access is granted narrowly and per task, every action is logged, and irreversible operations require a human to confirm them. What leaves the perimeter and what never does is your call.
Start with a single process — the most time-consuming and the most mechanical one. Sometimes it has to be straightened out first: automating a mess only buys you a faster mess. I will say that too, up to "AI is not needed here". The result shows up in weeks, not a quarter.
Every project is set up for AI agents: architecture and rules written down beside the code, a set of project commands, MCP servers wired to the database, the browser and the test suite. Agents write code, run end-to-end scenarios and dig through logs — under human supervision and mandatory review.
Project rules, architecture and decisions sit beside the code and evolve with it, instead of getting lost in chat threads.
The agent queries the database, drives a real browser and runs the tests. Validating an idea takes minutes, not half a day.
An agent reviews the code first, then a person. The automated pass catches the mechanical — missed checks, leaks, inconsistency — while the human looks at intent and consequences. Nothing reaches production without that second pass.
In-house solution
Working with agents seriously called for a harness of its own. It launches sessions and keeps each one in a separate process with a pseudoconsole, so a session survives both a restart and an update of the solution itself. The terminal is streamed to the browser over WebSocket, alongside a file browser with git diffs, push notifications and one-click updates. The interface is cross-platform: work continues from a laptop, a tablet or a phone, anywhere on the network.
In practice this cuts routine work several times over and lets one engineer carry a scope that used to require a team. The same experience goes into client products whenever AI has to live inside the system rather than sit next to it.
2019 — now
SDM Technical Lead
Kazakhstan
2019 — now
Treemple Senior Software Developer, co-founder
Canada
2013 — now
Kazakhstan Prime Innovation Architect
Kazakhstan
2015—2019
Monithor Inc. Senior Software Developer
Canada
2006—2013
NAT Kazakhstan Senior Software Developer, Technical Leader
Kazakhstan
A
From requirements to release and support. I own the outcome end to end, infrastructure and deployment included.
B
Architecture, code review, standards, hiring and growing developers. I join an existing team and take the engineering side.
C
A single session or a few: working through a specific question, a second opinion on architecture or stack choice, an assessment of a vendor proposal, help making a technical call. No long contract — you pay for the time and leave with an answer.
D
A hard look at the system you inherited, an honest assessment, a step-by-step migration plan, and the execution of it without downtime.
Send a couple of paragraphs about the problem — I reply within a business day with a suggestion on where to start. The first conversation commits you to nothing.