Zbyněk Lazárek — Engineering Dossier

§01

Premise

The hard part of building with AI agents isn't generating code. It's making a team of them reliable enough for production. I built the system that does it — and I have two live businesses to show for it.

The usual divide — I

Senior architects.

Can design sophisticated systems — and deliver at one person's pace.

The usual divide — II

AI tinkerers.

Can produce a dazzling demo — and never reach production reliability.

Where I work

Production-grade agentic delivery.

I architect the system, then run a governed team of AI agents — a CTO, engineers, QA, DevOps — against it, with deploy gates, persistent memory, documentation-first discipline and escalation built in. I own the architecture and the calls, and I work hand-in-hand with the people who make software real: domain experts who know the business, testers who break it, and stakeholders who set the priorities. The AI is the leverage; the team makes it land.

§02

Selected work — two case files

Not prototypes. Both are live, with real users and real money riding on them.

National deposit-return system

Enterprise infrastructure · client confidential

In
Production

A clean domain-driven rewrite of a live, country-scale recyclable-collection and deposit-return platform — migrated off a legacy monolith with the Strangler-Fig pattern, without disrupting production. Legacy events bridge into the new backend; bounded contexts take over one at a time; every cutover is gated and reversible.

Bounded contexts: 12
Backend C#: ~118,000 lines
Test files: 536
Sustained velocity: ~7.7 commits / day
Status: Live · national scale

Built with .NET 9 · Marten event sourcing · MassTransit · PostgreSQL · Angular · Azure AKS / GitOps · transactional outbox · App Insights

Anglická Výzva

English-learning SaaS · consumer subscription product

Live ·
Paying

A full-lifecycle consumer product — free trial to paid subscription, billing, gamified learning — on an event-sourced backend with mobile and admin frontends and end-to-end test coverage. Plus the part nobody sees: a zero-downtime tenant cutover and a bulk migration of 3,160 users and 33M points, fully idempotent.

Bounded contexts: 8
Passing tests: ~2,177
Architecture decision records: 26
Cutover downtime: Zero

Built with .NET 9 · Marten · CQRS · MassTransit · Angular · Playwright · Azure · passwordless auth · time-travel testing

§03

Competencies

a.

Software architecture

Domain-driven design, event sourcing, CQRS, modular monoliths with clean context isolation, Strangler-Fig migration, transactional-outbox correctness — decisions written down as 30+ ADRs.

b.

Agentic systems & AI engineering

Designing and running multi-agent AI orgs: role-scoped agents, persistent memory, an inter-agent message bus, governance and deploy gates, a monitoring dashboard, and bespoke orchestration tooling.

c.

Full-stack, full-lifecycle delivery

.NET / C# backends, Angular / TypeScript frontends, Playwright E2E, PostgreSQL, Docker, Azure AKS, GitOps, CI/CD, telemetry — documentation-first through to production operations. I own the whole pipeline.

d.

Engineering leadership as a system

I run the equivalent of a 9–12 person org as machinery: quality gates, code review, E2E gates, decision escalation, and a loop that turns every mistake into a written policy.

e.

Production operations

Zero-downtime cutovers, large data migrations, live incident diagnosis from telemetry, replay and ledger systems, deployment runbooks. I ship — and I keep it running.

f.

Knowledge sharing & enablement

I genuinely enjoy teaching this. Workshops and education calls on AI-leveraged delivery, agent orchestration and clean architecture — helping teams put agents to real production work, not just demos.

§04

Method — the operating system

An operating system for AI teams.

These systems reach production because of governance, not autocomplete. I built an OS that makes a team of AI agents behave like a disciplined engineering org — with the right people in the loop exactly where it counts: me on architecture, domain experts on correctness, testers on proof.

→Role-scoped agents — CEO, CTO, engineers, QA, DevOps, support
→Persistent memory — knowledge and lessons survive every session
→Quality gates — docs-first, code review, mandatory E2E before deploy
→Owner-only deploy — no agent ships to production unsupervised
→Decision escalation — every architectural call routed to the human
→Humans where it matters — domain experts, testers and stakeholders review the real decisions
→Portable — the whole org OS transfers between products by design

§05

Stack & colophon

Languages C# / .NET 9 · TypeScript
Backend Marten (event store) · MassTransit · PostgreSQL · MediatR
Frontend Angular · Tailwind CSS · Playwright
Platform Docker · Azure AKS / Kubernetes · GitOps · Kustomize
Observability App Insights · OpenTelemetry
Identity Azure AD B2C · JWT
Orchestration Claude · custom multi-agent tooling

§06

Correspondence

Building something that has to ship — fast, and for real?

I'm open to AI-native engineering leadership, principal / architect roles, and fractional-CTO work — anywhere AI-leveraged delivery is the point rather than a buzzword. I also run workshops and education calls, and I'm always glad to share what I've learned.

zbynek@lazarek.dev

lazarek.dev LinkedIn

On how the work gets done. AI agents give me the throughput of a much larger team — but production software comes from people too. I architect and decide; domain experts, testers and business stakeholders shape it, stress-test it and keep it honest. I lead the system; I'm not an island.