Od dziś · AI Practitioner & Builder

I help dev teams adopt AI in real projects — not just talk about it

10+ years of software engineering. I assess workflows, identify AI opportunities, design and ship working tools — with developers, not for them.

Assess development workflows and identify high-value AI opportunities
Design and prototype AI-assisted solutions using LLMs, prompt engineering and agentic workflows
Ship production-ready tools with CI/CD, monitoring and responsible AI practices
Run workshops and live coding sessions to drive team adoption
Claude APIOpenAI APIPrompt Engineering Agentic WorkflowsPython/FlaskNext.js DockerGitHub CopilotKotlin/Android
See AI projects ↓ Get in touch
Maciej Krawczyk
400+
developers reached
10+ years software engineering
2 AI products shipped & live in production
80% less time on content workflows — validated with real users
4× public talks on AI adoption for dev teams
Projects

AI in practice, not in theory

PlannerMag
01 / 05

PlannerMag

plannermag.io — Founder & Developer · Live product
Context
Solo founder, SaaS product — social media content planning for businesses
My role
Full workflow analysis, solution design, end-to-end implementation and production deployment
AI implementation
Agentic pipeline: Haiku for data collection, Sonnet for content generation, Leonardo.ai for image generation. Prompt engineering tuned per industry vertical.
Production setup
CI/CD, unit tests, code coverage, vulnerability checks, Sentry, uptime monitoring
Outcome: weekly content planning reduced from 3–4 hours to 15–30 minutes (~80% reduction). Validated with real paying users across multiple industries.
Challenge: initial LLM outputs were generic and inconsistent across industries. Solved through iterative prompt engineering and industry-specific context injection — outputs are now contextually relevant and consistent.
Claude APIAgentic workflowPrompt engineeringNext.jsPython/FlaskDocker
Open app →
IP BOX Studio
02 / 05

IP BOX Studio

AI-enabled workflow tool for tax compliance — built with GitHub Copilot
Context
Internal tool for developers filing IP Box tax returns in Poland — a manual, error-prone annual process
My role
Mapped the end-to-end compliance process, identified automation opportunities, designed and built the solution
AI implementation
Specification-first development with GitHub Copilot. Rule-based cost classification engine with YAML overrides, built iteratively with AI as a development partner.
Where AI made the difference
Without AI: 4h build time would have been 2–3 days. The tool itself wouldn't exist — not faster, simply wouldn't have been built.
Outcome: annual compliance task reduced from 7–9 hours to 1 hour. ROI achieved in year one. Tested across two full tax years — works reliably without modifications.
PythonFlaskpdfplumberAlpine.jsDockerGitHub Copilot
Speaking at Java Users Group
03 / 05

AI Adoption Talks — 400+ developers

IDEMIA internal · JUG Łódź · JUG Toruń · Zasmakuj pracy w Łodzi
Audience
Backend, frontend and mobile engineers — across 4 events, 400+ developers total
Format
Live coding with AI tools, real case studies, Q&A, pair-working demonstrations
What I showed
Vibe coding vs agentic coding — how to choose. Real failures (hallucinations, non-existent libraries, empty ZIP files) alongside real wins. PlannerMag built end-to-end with AI as a live case study. How to integrate AI into daily dev workflow without losing control.
IDEMIA internal: 300–400 attendees, English · JUG Łódź: ~40 developers · JUG Toruń: ~50 developers, recorded · Zasmakuj pracy: 50 students + teachers. Consistently highly rated in post-event surveys.
Live codingAI adoptionKnowledge sharingJUG ToruńJUG Łódź
04 / 05

Enterprise AI Adoption Framework

Internal AI Guild · 2024 · 2 months
My role
Architecture design, use case identification across teams, management presentations
Framework built
Discovery → opportunity scoring → risk assessment → phased rollout plan
What we delivered
AWS Bedrock infrastructure design, self-hosted LiteLLM gateway, team-based LLM access controls with cost tracking. Complete — never deployed.
Blocked at management level, not technical. Key outcome: deep understanding of how to assess AI readiness, score opportunities, and present the business case to leadership.
AWS BedrockLiteLLMAI governanceChange management
"The biggest barrier to AI adoption isn't the technology. It's the organisation."
— conclusion after 2 months of AI guild work, now applied when advising teams on adoption strategy
Poly — Voyager Free 60 Series Product Launch Award
05 / 05

Poly (HP)

Senior Android Developer / Technical Leader · 2022–2023
Context
Android app for Voyager headset series — joined what appeared to be a Greenfield project
My role
Technical Leader, Scrum Master, de-facto Product Owner — 4-person team
Real situation
Rescue mission: inherited a monolithic codebase split from a failed Flutter attempt — no tests, no DI, no modularisation, resistant team
What I did
Pushed for architectural refactoring: modularisation, dependency injection, unit + UI tests, CI/CD from scratch. Aligned team around best practices while maintaining delivery pace.
Delivered: Voyager Free 60 Series product launch on time. Recognised by Poly with a personal award for contribution to the product launch.
Technical LeadershipArchitectureModularisationCI/CDScrum MasterAndroid
Process

How I work as an AI Practitioner

01

Assess workflows

Identify repetitive, high-volume or error-prone steps in development and delivery workflows. Score opportunities by impact vs. implementation effort.

02

Design & prototype

Select the right AI approach — LLM, prompt chain, agentic workflow or simple automation. Build a working proof of concept with the team, not in isolation.

03

Ship responsibly

Production-ready means: validators on LLM outputs, PII handling, monitoring, fallbacks. AI as a tool — not a black box no one understands.

04

Drive adoption

Workshops, live coding sessions, pair-working with developers. The goal: teams that use AI every day because they understand it — not because they were told to.

How I build

AI-assisted development, done responsibly

🧩

Specification-first prompting

Every project starts with a written spec before any code is generated. AI asks clarifying questions, breaking the problem into atomic tasks.

Used in PlannerMag to keep prompts maintainable as requirements evolved across different industries.
🔁

Checkpoint-based development

Dense Git commits after every working step. When AI drifts or hallucinates — git reset --hard. Bold but not reckless.

Used in IP BOX Studio to keep scope contained and allow safe rollback during iterative development.
🛡️

Responsible AI by default

PII anonymised at input layer, validators on all LLM outputs, monitoring in production. For sensitive data: local LLM for pre/post anonymisation.

Applied in enterprise AI framework: data leakage checks, output evals, hallucination mitigation built into the architecture from day one.

Let's talk

Looking for someone who understands both engineering and AI adoption in dev teams?

LinkedIn → GitHub YouTube