Home Blog About MeContact
AI Engineer · Senior Developer

Coder.
AI Engineer.

15 years in Big Tech. I design AI systems, advise companies, and train teams on the tools I use myself every day.

Get in touchRead the blog
Aga Czyżewska
Engineering & AI

15 years of engineering, now building AI systems

Most recently applied to production AI: multi-agent workflows, local models, and privacy-first architecture. I lead teams, own quality end to end, and ship things that work — not demos.

Production AI Systems

Beyond simple prompts: multi-agent workflows and custom model integrations, architected to solve real operational problems.

Local & Private Models

Running models on my own infrastructure — vision (VLM), speech synthesis (XTTS), image generation (Flux.1) — without depending on a single vendor.

Security & Compliance

I know which data can go to the cloud and which must stay local, and I configure AI systems with GDPR and data protection built in from day one.

Team Leadership

Managing engineering behind browser extensions used by 100M+ people, with over a decade in test automation and quality systems. I mentor engineers and teach at conferences and workshops.

Tools

I know these from the inside — I use them in production

I don't test these — I build working products with them every day.

AI Models & Chats

ClaudeGeminiChatGPT / CodexPerplexity

Platforms & APIs

OpenRouterVertex AIAnthropic APIOllama

Local Models

XTTS v2WhisperQWenFlux.1and many more
Open Source & Side Projects

Building with AI

Production-grade side projects, built solo, full-stack, open to explore.

EtA — EPUB to Audiobook

Python CLI that converts EPUB files to audiobooks using XTTS v2 voice cloning. Handles chunking, preprocessing, FFmpeg postprocessing, and checkpoint resume. No cloud, no data upload.

GitHub

Autonomous Multi-Agent System

Multi-agent system for software engineering tasks in Big Tech — agents pick up tickets, gather context, write code, fix CI, and open PRs without human involvement. I build the orchestration layer and internal LLM integration.

AI English

AI English

PWA for professional English vocabulary. Spaced repetition algorithm + Gemini 2.0 Flash generating personalized word sets. React, TypeScript, Firebase, offline-capable.

Trinity

Family command center for tasks, calendar, and workouts. AI runs in the cloud, but a semantic layer ensures only what's necessary leaves the house — sensitive data stays private.

newsToPodcast

Daily feed from agents that learn what I care about — selected articles land as an audio podcast. Local TTS, my own voice, zero subscriptions. I read the news with my ears.

Garmin Trainer

Personal AI coach that analyses Garmin data — HRV, sleep, load — and plans a weekly training schedule matched to current fitness and running goals.

Aga Czyżewska
About Me

Aga Czyżewska — Developer, AI Engineer & Educator

I built my first computer at age 10 and created my first website at 15. For over 15 years I have been professionally navigating the IT landscape, evolving from a curious tinkerer to a strategic leader. I work independently, taking full initiative to deliver solutions that bridge the gap between complex engineering and real-world needs.

Currently, as a Senior JS Developer at eyeo (AdBlock Plus), I work on browser extensions used by over 100 million people worldwide. My focus on privacy solutions and high-scale performance requires a deep understanding of the modern web ecosystem and the responsibility of building for a global audience.

In the realm of Artificial Intelligence, I am a practitioner first. I've developed custom solutions including a local TTS-based audiobook creator and a fully automated AI-generated newsletter. I use Claude and advanced LLM workflows daily, not just to build software, but to refine the process of development itself and teach others how to master these tools.

I believe technology is only as valuable as our ability to share it. I train IT companies, run workshops for seniors at the University of the Third Age, and teach programming to youth in foster care through the Hakersi project. Whether presenting at international conferences or teaching cybersecurity in schools, my strength lies in translating complex technology into plain language for every age group.

15years of experience
100M+software users
Certificates & Education

Learning from the Best

AI_Devs

Elite, demanding AI development course — completed in the 1st cohort, 100% of tasks delivered on time.

Developer Jutra

Course covering architecture, product, and AI in practice for developers.

Web App Security

Niebezpiecznik certification, "Attack & Defense of Web Applications" — hands-on web security from both the attacker's and defender's side.

Defensive Programming

Niebezpiecznik certification in secure, defensive software design.

Mobile Security

Niebezpiecznik certification in mobile application security.

ISTQB Certified

ISTQB Foundation Level certificate in software testing.

Communication

Leaders Island training in professional communication.

Master's
Biomedical Engineering

Specialization: Informatics & Electronics

Master's
Quality Management

Production Engineering

How Can I Help You?

Skills built over 15 years in the industry

AI & LLM

Architecture, prompting, optimization. I help teams put AI to work effectively, in the business and in the classroom.

Tech Education

Workshops for youth, seniors, and IT teams. Practical skills you can apply immediately.

Test Automation

Over 10 years of experience. I build robust test frameworks that ensure quality.

Full-Stack Dev

JS specialist (React, Node, Express). Complete solutions from frontend to backend.

Deeper than a prompt

I don't just use AI — I understand how it works

Writing prompts in ChatGPT is the tip of the iceberg.
I'll show you the rest.

Does company data have to go to the cloud?
No — and that's one of the first things we establish. Some data can safely go to an external model's API. Other data should never leave the company. I have experience with three approaches: local models on your own machines (full privacy, higher infrastructure cost), a semantic layer that filters sensitive data before it reaches the AI provider, and dedicated cloud environments (e.g. Azure OpenAI, Vertex AI) with agreements guaranteeing your data won't be used for training. The right setup depends on scale, budget, and the nature of your data. I can walk you through each option.
Is AI deployment GDPR-compliant?
Not automatically — and that's the most common mistake. AI deployed out of the box without proper thought can violate GDPR. Every case is different: different data, different provider, different process. Every system I build is designed with privacy from the start — not bolted on at the end. That's not accidental: my background includes cybersecurity and data protection, so compliance is a natural part of the architecture for me, not a checkbox.
Can an AI agent work fully autonomously?
It depends on the scope. Small, well-defined tasks — yes, an agent handles them autonomously. Complex processes — I typically start with a supervised mode where a human approves key steps before the system moves to full automation. Important: "AI" is often not just the model, but the harness — scripts, guardrails, retry mechanisms, and validation logic surrounding the model. These determine reliability. Human-in-the-loop isn't a weakness — it's a deliberate architectural decision.
How do language models actually work?
In simple terms: a language model is a massive statistical machine, trained on billions of texts, that calculates which word most commonly followed a given context — and picks that word as the next one. It sounds trivial, but the scale makes the results remarkable.

Technically: it's a neural network (transformer architecture) that processes text as tokens — fragments of words, not individual letters. Each token is converted to an embedding — a vector of numbers describing its meaning. The attention mechanism lets the model weigh which parts of the context matter for the next token.

This is where both the capabilities and the limitations come from: the model doesn't "remember" — it sees a context window. It doesn't "think" — it predicts the next token. Temperature controls the randomness of those predictions. That's why the same question gives different answers.

Hallucinations aren't a bug — they're a property of a model that optimizes for text coherence, not factual accuracy. Knowing this, I know where to put guardrails and how to design systems that don't fail in production.
AI is new. How do I know you're not just another LinkedIn "AI guru"?
Fair question — there are a lot of those right now. The difference is simple: I write code that goes to production. 15 years as a developer — MES systems for L'Oréal, browser extensions used by over 100 million people at Big Tech, publications on the Google Chrome Developers Blog. I'm a developer who has shipped code at scale.

At my company I'm the person responsible for driving AI adoption across engineering teams — building AI-powered solutions every single day.

My systems are already running in production. I'm AI Native — I don't test, I don't "explore the potential", I just build.

A LinkedIn guru won't explain what risks a given solution carries — because they've never actually built one.
Can AI integrate with the tools we already use?
Yes — and that's one of the first things I look at. There's no point building from scratch if AI can connect to an existing stack: CRM, helpdesk, Slack, internal knowledge base, Jira. The MCP protocol (Model Context Protocol) lets models securely read and write data in your tools without having to pipe everything into one system. I build and deploy these MCP servers in production.
What can go wrong when AI has access to company data?
Quite a lot — which is why limiting agent permissions is a core part of my architecture. An agent with network access can accidentally exfiltrate data. An agent with write access can overwrite records. The model can make a mistake and nobody notices for a week. I design systems with the principle of least privilege, approval mechanisms for critical actions, and logging that catches anomalies before they become incidents.
Aga Czyżewska
Contact

Let's talk

Have a question, a project, or want to discuss working together? Write to me directly.

Aga Czyżewska
czesc@czyzewska.pro
Send me a message

I've worked with

HappydemicsTest automation workshops for dev teamScandiSpaAI consulting & websiteDietetyk z PodhalaWebsite & online visibilityŚwarne ButyTools audit & IT setupHappydemicsTest automation workshops for dev teamScandiSpaAI consulting & websiteDietetyk z PodhalaWebsite & online visibilityŚwarne ButyTools audit & IT setup

Navigation

HomeBlog

Information

Privacy Policy

 

Made in Nowy Targ, Poland!
© Agnieszka Czyżewska

Polski