Andriy ZubetsSenior Full-Stack & AI Engineer

Senior Full-Stack & AI Engineer

I build AI products that work in production and pay for themselves.

Technical owner of a multi-service AI SaaS: Laravel product, Python/FastAPI AI services, document pipelines on Gemini, AI economics designed in from day one. I decide the architecture, AI agents write most of the code, and I own the result.

See the flagship case LinkedIn

Auvergne-Rhône-Alpes, France · Remote across Europe

Andriy Zubets
9+years building software
40+systems shipped
10M+records in a paid data product
3countries in production

Selected work

Systems built for real work

The projects below show the problems I take on: unclear requirements, messy data, many integrations, and a product that has to work every day.

Flagship · 2024–now

Multi-service AI SaaS for document-heavy decisions

The platform ingests large volumes of multilingual documents from public sources, converts PDF, DOCX and XLSX, finds what matters with LLMs, and drafts structured answers for users. Every AI action is priced before it runs.

≈1Mrecords ingested per year
3countries in production
4services and repositories
01CollectPublic sources, 3 countries
02PrepareIntake and processing
03ConvertPDF, DOCX and XLSX
04AnalyzeLLM pipelines on Gemini
05UseAnswers and workflows

My responsibility

  • Turn stakeholder ideas into clear features and user flows
  • Make the main architecture, database, API, and delivery decisions
  • Own deployment, observability, and ongoing improvement

Decisions that paid off

  • Split into services only when one application stopped being enough
  • Price every AI action before it runs, so the product stays profitable at scale
  • Benchmark document tools and route each file through the best one
  • Run full browser-level user journeys before every release
Laravel 11PHP 8.2PostgreSQLRedisPython / FastAPI servicesPlaywrightDockerAzure App ServiceVertex AI / GeminiClaudeOpenAIStripe

More examples

Three other products

Different problems, each delivered for real use.

Human-in-the-loop AI pipeline2,000+articles

Editorial automation

A publishing process that keeps people in control while AI handles the repetitive work.

  • Generate articles and images, then wait for human approval
  • Translate approved content into two languages
  • Create supporting content, publish, schedule, and submit for indexing
Paid data product10M+records

Automotive intelligence

Several large datasets became one fast search and monitoring product. Served paying subscribers until 2022.

  • Search, saved records, filters, change tracking, and alerts
  • Subscriptions, payments, and multi-factor authentication
  • VIN decoding and tools for administrators
Raw data to product15Ksource records

Venue search

A messy venue list became a structured search product with maps, filters, and useful pages.

  • Clean and check records, duplicates, locations, and images
  • Build map search and filters for different user needs
  • Generate 12,000+ venue pages and 9,000+ search pages

How I work

Agentic engineering, with an owner's judgment.

I do not start with a favourite technology. I start with how the business works and what success looks like, then decide what gets built.

01

Understand

Talk to the people involved and map the real workflow, not only the first request.

02

Decide

Check risks and edge cases, compare options, and remove complexity that brings no value.

03

Build

Turn the agreed plan into software, integrations, data flows, and a usable interface.

04

Verify

Test real user journeys, find what breaks before users do, deploy, and keep improving.

AI writes most of the code. I own every decision.

I own

Problem framing, constraints and edge cases, architecture and data model, product behaviour, and the production result.

Agents do

Implementation, tests, debugging, research, and cross-model review. Claude Code, Codex, and specialised agents, up to ~20 in parallel.

Experience

Business experience first. Software for the last 9+ years.

I moved from managing operations into building software. That is why I naturally think about users, cost, risk, and the result—not only the code.

  1. Business and operations

    Managed teams, costs, risk, and profitability in organizations of up to 300 people.

  2. Frontend development

    Started by turning designs into responsive websites.

  3. Full-stack systems

    Built databases, APIs, integrations, search, payments, automation, and complete business tools.

  4. AI products, agentic engineering

    Technical owner of a multi-service AI SaaS: LLM pipelines, AI economics, and the testing strategy for the team.

About

I understand the business side too.

Before software, I spent years managing people and operations. Today that helps me challenge a feature before a team spends time building it.

Will people use it? What breaks first? Is the extra complexity worth it? What does the business really need?
Auvergne-Rhône-Alpes, FranceEnglish: professionalUkrainian, Russian: native

Technical toolbox

Tools I use to deliver products

Core

  • PHP
  • Laravel
  • Python / FastAPI
  • JavaScript
  • PostgreSQL
  • MySQL
  • REST APIs
  • Application architecture

Production work

  • Redis
  • Docker
  • Playwright
  • Azure App Service
  • Google Cloud
  • Stripe

AI systems

  • Gemini
  • OpenAI
  • Anthropic
  • LLM pipelines
  • AI agents
  • Claude Code & Codex (agentic workflow)
  • Structured outputs
  • Vector search
  • Prompt evaluation
  • AI economics

Contact

Building something where AI has to actually work?

Happy to talk about AI products, architecture and agentic engineering. Auvergne-Rhône-Alpes, France · remote across Europe.

LinkedIn Email Telegram WhatsApp