AI engineer · architect · tooling builder

Context before LLM.

AI agents · multi-tenant systems · test automation

AI tooling, RAG deployments, QA training - production-ready, AGPL distillates included. 18 years shipping software, last 12 months building AI agents that compound.

30 min · free · no slides · usually replies within 24h

18+
years in IT · since 2008
19
specialized AI agents
23
custom skills · pipelines
34K
LOC production QA
9
production systems · CDAT
9
packages · public AGPL toolkit

services

Three things I do, end-to-end.

AI tooling, context engineering, QA workflow. Each service rooted in real production work - source code, audit logs, and AGPL distillates where it makes sense. No black box - receipts included.

[01] · AI-TOOLING

AI Engineer · agents that audit and build

AI tooling for QA, a11y, and AI-native discoverability. Custom agents on demand, flagship audits ready to run.


  • WCAG 2.2 AA audits - flagship · Grade A in days · 5 specialists + Lead · Series #01
  • AI-SEO Pack - llms.txt · /mcp · JSON-LD · per-article OG · public release soon
  • Custom AI agents - your stack, your workflow · healers, pipelines, copilots

~5-14 days · scope on intro call

[02] · CONTEXT-ENGINEERING

Context Engineering · context before LLM

Don't blast AI everywhere. Prepare deterministic context first - for agents and teams.


  • jarvis-brain - multi-tenant federated graph · MCP server · auto-update on git push
  • Deterministic pipelines - Figma · Jira · git · clean data BEFORE the LLM sees it
  • Stack - Qdrant + Ollama + MCP · multi-tenant · 600+ tests · live at brain.sdet.it

~2-4 weeks · scope on intro call

[03] · QA-ENGINEERING

QA Engineering · workflow + training

Your QA team ships AI-augmented tests that survive sprint 1 - with the dispatcher to run them at scale.


  • CDAT pattern - Components · Data · Actions · Tests · MIT · battle-tested in production
  • Multi-agent dispatcher - Claude Code & OpenCode · model-agnostic · 100-200 tests in 2-3 days
  • Paired engineering - hands-on Playwright + AI · 2-day sprints with your team

training ~3-5 days · setup ~1-2 weeks

how we work

Four steps. No surprises.

Diagnostic before engagement. Fixed scope where it makes sense, T&M where it doesn't. Receipts every Friday.

01

Intro call

30 minutes. You describe the pain. I tell you whether I'm the right person. No slides.

free · 30 min

02

Diagnostic

Paid pilot, scoped deliverable. Audit, prototype, or evidence of feasibility - your choice.

paid · 3-5 days

03

Engagement

Fixed scope or T&M. Weekly demos, async-first communication, shared decision log.

scoped · 2-8 weeks

04

Handoff

Docs, recorded walkthroughs, training. 30-day support window included by default.

docs + 30 days support

stack

Tools I actually use.

Battle-tested across portfolio.sdet.it, cdat.sdet.it, brain.sdet.it, and 9 production systems. Open-source by default, AGPL where it matters.

lang/ TypeScript lang/ Python fw/ Astro fw/ Nuxt 3 fw/ Vue 3 test/ Playwright a11y/ axe-core ai/ Claude Code ai/ Codex ai/ OpenCode ai/ Cursor ai/ Ollama ai/ MLX ai/ MCP db/ Qdrant infra/ Docker infra/ n8n lic/ AGPL-3.0

from the field

Field notes, not testimonials.

Excerpts from public case studies. Each links to the full article with numbers, commits, and what didn't work.

F → A in 8 commits, 75 minutes total CC work. Real audit on a real production site. Source code linked.

Static regex saw 0 real findings. Dynamic axe saw 3. AI agents reading source saw 18. Same project, different views.

5% auto-fix. Humans handle the rest. No magic, no marketing - the boring math of accessibility remediation.

// CONTACT

Stop guessing about your test infra.

30 minutes, no slides, no NDA dance. You describe the pain. I tell you if I can help.

Usually replies within 24h · Europe/Warsaw timezone