Company — About
Humans. In the Loop.
ZemiData is a data and AI consultancy and a maker of data and AI products. We design, build, and operate the data pipelines, machine learning models, and LLM-powered applications that products depend on — and we stay involved once those systems are live, where most of the real work actually happens.
Our name is also our principle: Humans. In the Loop. We believe automated systems should make people more capable, not less accountable. So we build for visibility and human judgment — systems an operator can actually see into, measure, and trust, rather than black boxes that quietly drift out from under you.
What we believe
Four things show up in everything we ship:
- You should be able to see into it. Observable, measurable, accountable — every decision can explain itself. No black boxes, no “trust us.”
- You should be able to keep it. Self-hosted, open where it can be, no lock-in. Protection and control without shipping your data to someone else’s cloud to be trusted.
- Trust is earned, not asked for. We run what we sell before we sell it. Where a tool can only attest to something rather than enforce it, it says so.
- The work carries something forward. We build things meant to outlast the engagement — and the people who built them.
Bigger than any one of us
ZemiData reflects the people who build it, but it was never meant to be about them. It’s a culmination — a career’s worth of building data systems that have to survive contact with the real world — and a continuation of a tradition of careful, made things. The company is named for a cemí, a carved object that holds meaning; its products carry Taíno names, each a small story about what the thing does. That’s deliberate: the work is supposed to carry something.
The aim is to build instruments other people run themselves — tools that serve everyone they touch, on both sides of the table — so the principle outlives any single engagement, and any single person.
Founder
ZemiData was founded by Jonathan Pulliza, a data engineering and analytics leader whose work spans cloud data platforms (Snowflake, Databricks, AWS, Azure), MLOps, search engineering, and consulting — with a doctorate focused on how people work with generative AI. He has built and run data systems for standards organizations, consultancies, and startups, and builds ZemiData’s products with the same hands that deliver its engagements.
How we work
We run what we sell. Every ZemiData product started as a problem inside a real engagement — a client who needed proof their AI hadn’t changed under them, a strategy decision too expensive to get wrong on instinct. We build the tool, run it on our own systems in production first, and ship it only when it has survived us. Consulting sharpens the products; the products make the consulting testable. The loop is the point.
Two examples, running right now:
- Behique — our agent-drift monitor — watches a live deployment of our own coaching application in production, so it has to catch real drift before we’d ever point it at a client’s system.
- Vejigante — our data-access policy compiler — runs against that same application’s Postgres database, planning and explaining every access change on data we actually depend on.
We don’t recommend an instrument we haven’t already trusted with our own work.