A. Volosznyikov Portfolio ’26
ENDE Open to AI roles --:--
AI Systems Engineer Portfolio — 2026 N° 01

I build the parts you never have to think about.1

Index

A developer gone deep on AI. At home I serve open-weight models from my own small cluster, runbooked like production: the limit of my reach, not of my ambition. I want this work at a scale my spare room can’t hold.

Index Six sections, one page
  1. 01PositionThe pivot, stated plainly
  2. 02Selected WorkAI systems, platforms behind them
  3. 03PracticeWhat I do, and how
  4. 04LogFourteen honest years
  5. 05CompanyThe crew, so far
  6. 06ContactStart a conversation
01 — Position Why this site looks like this

Good infrastructure is invisible. No red dashboards, no heroics at 3 a.m. Just systems that hold their shape while everything around them moves. I design for that silence.

Fourteen years from freelance PHP to production systems for Swiss enterprise. I’m a doer by nature and a workhorse by reputation. Happiest where the work is concrete and the system has to hold. Born in Ukraine, raised in Hungary, at home in Switzerland.

The architect draws the arch; I’m the one who knows when the mortar is just right and how the brick will settle. What I want now is simple: AI engineering, paid, at a scale bigger than my spare room, with a team and a budget. I bring what AI teams usually learn the hard way: IaC, secret handling, CI/CD, incident calm, plus daily hands-on time with the inference stack most people only read about.

Focus
LLM inference · Agents · Azure
Base
Switzerland, on-site & remote · Working in English
Status
Senior engineer · Open to AI roles
Clearance
Swiss security-cleared
02 — Selected Work AI systems first, platforms behind them · Details on request2

012025 — NowLab · AI infrastructure

Home inference, production habits

Open-weight models (GLM, Qwen, DeepSeek) served through vLLM, from consumer GPUs up to Grace Blackwell. Dual GX10s are the best I could buy; the runbooks are written for the clusters I haven’t been handed yet.

vLLM · CUDA · Self-hosted

022026Lab · Agents

Self-hosted agent tooling

A DeepSeek Harness search-and-fetch plugin on self-hosted Firecrawl, built for my own bench. Nothing phones home: my day job lives in banking-grade closed systems, and my models live at home for the same reason.

TypeScript · dsh · Firecrawl · Self-hosted

032024 — NowCloud operations

Shareholder meetings that cannot fail

I run the Azure infrastructure, plus React and Spring Boot applications, behind annual general meetings: seasonal peaks, no tolerance for downtime. Security at both levels: hardened code and dependencies above, reviewed configurations, access, and secrets below.

Azure · Terraform · Spring Boot · React

042021 — 2024Re-platforming

Monolith to Azure, without the drama

Part of the team that moved a monolithic on-premises system to Azure. My share: legacy code containerized to run on VMs, behind NSGs and a WAF, patched and up to date.

Docker · Terraform · NSG · WAF

03 — Practice Four things, done well

Four things I do well. Everything else I delegate, automate, or decline.

LLM inference & serving
Open-weight models on my own GPU cluster, served for daily use: OpenAI-compatible APIs, metrics, gateways, and model lifecycle from download to benchmark. Profiling, optimizing, finding the bottleneck. I know this stack from operating it. Not just from calling it.
Agents & evals
Harnesses, tool integrations, and two-model orchestration against local models, including self-hosted search tooling. Artifact pipelines with compile gates. Model output is untrusted input until proven otherwise.
Cloud & platform
Azure platforms with Terraform and Ansible that behave the same on Friday as on Monday. Financial-grade, audit-friendly. The boring foundation AI systems need underneath them.
DevSecOps
Dependency management, secret handling, access reviews, configuration hardening, for the application and the infrastructure under it. AI systems get the same treatment: no keys in prompts, no trust without verification.

Operating principles

  1. Boring technology, operated brilliantly.
  2. If it pages at night, the design failed first.
  3. Automate the toil; keep a human for the judgment.
  4. Invisible is a feature. Ship silence.
  5. Execution outranks credentials.
  6. Live on the bleeding edge; find the bottleneck.
04 — Log A short ledger
PeriodRoleContextGround covered
2025 — nowAI Systems, IndependentHomelab: self-hosted LLM servingvLLM · Evals · Agents
2024 — nowDevOps / DevSecOps EngineerShareholder meeting systems, AzureSecurity · Terraform · Spring Boot
2023 — 2024Project Lead, AutomationFinancial infrastructure, RHELAnsible · DRP · GitLab
2021 — 2024Cloud DevOps EngineerLegacy monolith → Azure VMsDocker · NSG · WAF
2020 — 2021Full-Stack Developer, DevOpsServerless meeting applicationsReact · Functions · AD B2C
2019 — 2020Project Lead, DeveloperSustainability investing appReact · Node · Jenkins
2018 — 2019DeveloperOnline identification serviceCloud apps · Onshore–offshore
2016 — 2018Frontend DeveloperTelecom modernizationSitebuilds · Design implementation
2012 — 2016Full-Stack DeveloperFreelance & agency workPHP · SysAdmin

Fourteen years of production, from freelance PHP to AI systems. Full CV with client history and references on request.

05 — Company The crew, so far

I do my best work beside people better than me at what I’m not.

01Architect

The architect I build beside

Twenty-five years of mission-critical systems, most of it inside Swisscom, the deepest architect I have ever met, and a former coworker. He draws the arch; I lay the bricks.

Udanvita · Sovereign AI · Switzerland

Notes

1Boring on purpose. In infrastructure, excitement is usually an incident with a timeline.

2Client systems can’t always be shown. I’m happy to walk through architecture and decisions on a call.

3This page ships no framework, no images, and no trackers. Designed and built with an AI assistant. The right tool for the task; my local models handle the half that stays private.

06 — Contact

Have something that needs to hold?

I’m looking for one well-scoped AI system to pour fourteen years of production discipline into: inference, agents, or the platform underneath. Tell me what you’re building; I’ll tell you honestly whether I’m the right pair of hands. And if it’s something I love, you’ll get the cherry on the cake too.

hello@volosz.com