I’m Krzysztof K. Zdeb (a.k.a. KKZ) - CTO at BitPeak, a Data & AI consultancy. I’ve been programming since ~1988 and working professionally in the data domain since 2003 across various industries (telecom, banking, aviation, public administration, pharma, retail, ...). And I still design and implement the systems I write about, which is where most of this comes from.
What this is about
The question I keep coming back to nowadays: how do you get an AI system to act in the interest of one specific company?
Data alone won’t do it. The system also has to understand what the company’s terms mean, what it is trying to achieve, who may decide what, and which limits it must respect.
The chain runs:
data → semantics → knowledge → reasoning → judgment → action
Most enterprise AI work stops at knowledge and calls it done. It skips the following steps: choosing what to do (and why), for this company, within its goals and its limits. Most writing on AI says to leave that step to humans. I have findings on that. They don’t fit in a paragraph, which is the reason this publication exists.
Every piece here sits on one link of that chain. Some sit at the bottom: data engineering, ML/AI engineering, data platforms, data products, and the unglamorous work of keeping them running. Nothing above can run reliably without them. Others sit higher: semantics and ontologies, agents in production, and how AI is changing the way we build and deliver.
What you’ll find here
Long pieces, written from systems that ran in production, and sometimes from ones that didn’t. Working notes when an idea is still forming; I’ll say when it is. The occasional teardown of a method or a tool that gets more credit than it earns.
On the technology side, expect mainly open-source tooling and Databricks.
Who it’s for
Architects, data and platform leads, and CTOs who have to make AI work inside a real company, with its own vocabulary, its own rules, and the people who own both.
And, of course, fellow data & AI geeks.
Subscribe
New pieces arrive by email, and that is the whole arrangement. If you’d rather read in an app, Substack has one.
Follow along, and let me know where we agree and where I'm wrong… 🤔
