About · Analytics engineering · B2B SaaS

Trustworthy data.
Useful AI.

I’m Piotr, an analytics engineer building dependable data foundations and exploring AI agents grounded in semantic layers—with context, accuracy, evaluation, and cost treated as first-class design concerns.

Start with a clear definition. Then make it useful.

Good analytics starts before the dashboard. It starts with shared definitions, well-modelled data, checks that catch meaningful failures, and a clear path from a question to a trusted answer.

That foundation is also what makes AI useful in analytics. I’m especially interested in agents that work from governed semantic layers, carry the right context, ask when a question is ambiguous, and can be evaluated against known expectations.

My writing and open-source projects explore these ideas with synthetic examples and reusable engineering patterns.

01
Define before you automate

An agent can’t repair an unclear metric definition by sounding confident.

02
Test what matters

Prefer checks that protect business meaning, not just pipeline mechanics.

03
Make the system inspectable

Keep context, tools, evaluation, and operating cost visible to the team.

Reliable systems, from model to answer.

Practical work across analytics engineering and governed AI.

01 · Foundations

Metrics people can agree on

Model shared definitions and trusted data products that work across reporting, teams, and tools.

dbt · Snowflake · data quality

02 · Applications

AI grounded in business meaning

Explore agent patterns that use governed metrics and explicit context rather than inventing answers.

Semantic layers · context · tools

03 · Reliability

Evidence before confidence

Use evaluation, deterministic checks, and cost awareness to understand whether an agent is helping.

Accuracy · judges · cost

The current toolkit.

Analytics engineering

dbtSnowflakeDuckDBSQL

Agent foundations

Semantic layersTool designContextMemory

Evaluation

Golden setsDeterministic checksLLM judgesCost controls

Engineering

PythonGitHub ActionsStreamlit

Have a thoughtful data or AI problem?

I’m always glad to compare notes on analytics engineering, semantic layers, or building more dependable AI agents.

Email me