An engineering artel · software & AI systems
Artel (арте́ль) is an old word for a self-governed crew of craftsmen who share the work and the responsibility. That's us: a small team of engineers, no middle managers. We take a project whole — from idea to a system running in production — and each of us signs the result with their own name.
Assistants for your customers and your staff: they answer from your data, plug into your systems and take action — filing tickets, parsing documents, preparing reports. Running inside your perimeter. We don't "adopt AI" — we automate a specific process with a measurable result.
RAG · agents · integrations · self-hosted
We design and build what a business runs on: customer portals, billing and payments, APIs and integrations with external systems. From an MVP in a month to multi-tenant platforms under load — with an architecture that survives growth instead of being rewritten for it.
TypeScript · React · Go · PostgreSQL
We make releases daily and boring, and failures invisible to your users: CI/CD with one-command rollback, fault-tolerant DNS, monitoring that wakes us before your clients notice. Migrations between clouds or onto your own hardware — that's us too.
Kubernetes · Terraform · Grafana · PowerDNS
We don't abandon what we've built — and we adopt what others left behind: systems without a team, contractor inheritance, "it works but nobody knows how". We start with an audit, then support and improvements under a clear SLA, with on-call duty and a one-hour response — not "a manager will get back to you".
SLA · on-call · audits · 1-hour response
You discuss the problem with the person who writes the code and makes the technical calls. No message loss through three layers of account managers.
Before work starts we put the scope, deadline and price in writing. If something changes along the way, we change the agreement explicitly — we don't quietly inflate the invoice.
All code is reviewed inside the artel, and we fix our own bugs at our own expense. The artel's reputation is collective — nobody lets sloppy work through.
Source code, access, documentation and IP are yours from day one. Leaving us is easy; we'd rather you stay by choice, not by lock-in.
Foreman · generalist
24 years in IT, since age eighteen. A generalist: from architecture and AI to infrastructure and production. Leads projects from first call to handover.
Backend & data
Go, PostgreSQL and everything that has to hold under load. Happiest when p99 is under a hundred milliseconds.
Frontend & interfaces
TypeScript, React and a healthy obsession with pixels. Believes an interface has a duty to be fast.
LLM & RAG systems
Teaches models to work from the client's data: RAG, fine-tuning, evaluation. Trusts logs, not benchmarks.
Data & storage
PostgreSQL, ClickHouse, zero-loss migrations. Knows data outlives any code — and treats it accordingly.
Backend & load
Go, Rust, queues and caches. Happiest when the system stays bored under peak traffic.
Integrations & APIs
Banks, payment providers, government systems and other “their API is peculiar” cases. Connects what refused to connect.
Frontend & design systems
Components, tokens, accessibility. Believes a design system is a contract, not a library.
Per job
When a job needs a rare craft — mobile, design, security — we bring in trusted specialists we've worked with before. The artel vouches for them as for its own.
We talk through the problem and tell you honestly whether we'll take it. If it's not our craft, we'll point you to someone whose it is.
We draft the plan: milestones, deadlines, price, acceptance criteria. We sign — and from then on, the agreement is the law.
Working software every two weeks. You see progress in the product, not in status reports.
We ship to production and hand over documentation and access. After that — support under an SLA, if you want it.
A RAG bot over 40,000 regulations and waybills: operators now find answers in seconds instead of minutes.
An MVP portal for a services company: plans, invoices, tickets. From first call to first paying customer in 10 weeks.
Multi-provider DNS architecture, CI/CD and monitoring: releases became daily, downtime over six months — zero.
A unified platform for financial operations: accounts and payments, bank integrations, role-based access and a full audit trail for every operation. Designed for scale and for regulators alike: reviews, certification and compliance are part of the project — not a surprise after launch.
A watchman for multi-provider DNS: compares SOA serials across your primary and every secondary, and raises a Telegram alarm on drift or zone expiry. Born inside our own infrastructure — now running for clients under our care.
request a demo →A boxed RAG assistant over your company's internal documentation: deploys inside your perimeter, never sends data out. The same engine we build into client projects — in a self-hosted edition.
ask for beta access →A protective screen between your users and LLMs: intercepts prompt injections, strips personal data before it reaches the model, enforces access policies and keeps a full request log. The product is built and running on our test bench; it hasn't been battle-tested in production yet — which is why first pilot deployments get special terms and the artel's full attention.
become a pilot →An actor-oriented DBMS with a built-in programming language: business logic and data live in one plane, with no "app ↔ database" gap. Logic is written in its own language, CDL (types, ownership, process-based concurrency); queries in SQL++, compiled through QIR. All code runs in strict isolation — security is built into the runtime, not bolted on top. Clustered architecture is supported out of the box.
follow the project →Honest answer: it depends on scope, and anyone quoting a price before understanding the problem is guessing. Rough anchors: a web-service MVP usually takes a pair of masters 4–6 weeks; an AI assistant over your data, 6–8. After the first call (which is free) we send a plan with a concrete price — and it doesn't change unless the agreement does.
Both. For well-bounded scope we prefer fixed price — that puts the risk on us, not you. For research-flavoured work and long-term care, T&M in two-week sprints is more honest: you see results every sprint and can stop at any point. We'll tell you which fits your job on the first call.
Yes — that's a fair share of our work. We start with an audit: we read the code and the infrastructure, then tell you plainly what's worth saving and what's cheaper to rebuild — it goes both ways. The audit is a small fixed-price step, and afterwards you're free to take the findings anywhere, even back to the previous team.
Yes — your template or ours. We're also fine with a no-portfolio clause: the artel's reputation is built on happy clients, not on logos used without asking.
We take it if the problem interests us or we have a free window. A small job is a good way to meet without big stakes: you see how we work, we see how we get along. Many long agreements started exactly like that.
First call and agreement — usually within a week of your message. The actual start depends on the artel's load: we give you a real date up front and never start "on paper" just to reserve you. That would be dishonest to the work.
Describe the problem in a couple of sentences — we'll come back within one business day with clarifying questions or, sometimes, straight with a plan. Looking at a problem and saying "we'll take it / we won't" costs nothing.
We reply within one business day