Skip to content
fewtokensai
About

Michał Czyżewski

AI Engineering Manager · AI consultant · founder of fewtokensai

AI engineer and team lead with 10 years of production work.

I build and lead AI Engineering teams — hiring, structure, career paths, on-call, code review, mentoring. I work directly with product, compliance and legal.

I design Model Context Protocol servers for production — with auth, audit log, idempotency, backpressure. I treat MCP as distributed-systems engineering, not a demo.

I build automated bookkeeping systems (AutoX) — OCR + NLP + LLM turning invoices into ready-to-post accounting entries at scale.

I ship AI decision assistants — like the PKD AI classifier matching a Polish business activity code from a free-text company description. Assistant, not a black box.

I ship LLMs into regulated environments — fintech, trading, financial services. Every layer goes through Compliance and Legal.

I build predictive models, scoring, churn analysis with measurable P&L impact. Classical ML still wins 80% of cases.


What I'm good at

LLM & GenAI: RAG, MCP, A2A, agent architectures, fine-tuning, evaluation frameworks.
Compliance: GDPR, EU AI Act, data residency, DPIA, audit logging.
Classical ML: client scoring, churn, predictive modeling, A/B testing, MLOps.
Data architecture: AWS (S3 + dbt + Redshift), GCP (BigQuery, Vertex AI), data lakehouse.
Stack: Python, SQL, LangChain, HuggingFace, Mojo, Docker, CI/CD.
Leadership: hiring, OKR, agile, mentoring, change management.

Background and community

AGH alumnus (Electrical Engineering), member of Mensa. Regularly read NeurIPS, ICML, ACL papers and attend conferences.

  • Applied ML Days, Lausanne — 2026
  • Applied ML Days, Lausanne — 2025
  • Applied ML Days, Lausanne — 2022
  • NeurIPS — 2021
  • Applied ML Days, Lausanne — 2020
  • PyData, Warsaw — 2019

Outside work — sailing, snowboarding, squash, Asian cooking, fantasy, RPGs. The best AI ideas usually arrive somewhere between the slope and the harbour. More personal: czyzewskim.pl (Polish only).

Let's talk about your AI

Let's talk.

30 minutes, no obligation. Tell me where your AI initiative is stuck or what you're planning — you'll leave with concrete next steps.