Yevgen Ponomarenko¶
I turn fragmented enterprise knowledge into governed, usable intelligence — semantic models, data products and agentic systems that operating teams can actually trust. And I lead the delivery that gets them shipped.
Seventeen years building enterprise data platforms, the last several spent on the harder half of the AI problem: not making a model answer, but making an organisation able to rely on the answer. Currently a Delivery Manager and hands-on AI engineer at EPAM.
How I work¶
Most enterprise AI failures are not model failures. They happen in the gap between data that exists and meaning an organisation agrees on. This is the loop I build for:
Skip any stage and you get a demo. Close the loop and you get a system a business can put weight on.
Featured¶
Proof · case study
A governed knowledge platform for revenue growth management¶
A global CPG manufacturer could not answer its own questions about its own numbers. Ingestion to grounded answer in roughly three months, with a measured quality baseline — and an honest account of what stood between the prototype and production.
Method · reusable
Fail loud in knowledge systems¶
An incomplete corpus that answers confidently is more dangerous than a pipeline that stops. Contract validation with quarantine, and why it was the highest-leverage decision in the build.
Inquiry · public source
A publishing platform for a family memoir¶
One source, four languages, three output formats, edge-gated access, and two licences in one repository. The same discipline as the enterprise work, applied where there is no budget and nobody to hand it over to.
What I am working on now¶
Three questions I am actively testing, rather than a topic list:
- When does a semantic layer need an owner rather than a document? Every knowledge platform I have built eventually stalls on the same thing — a definition nobody is accountable for. I am trying to work out the smallest governance structure that actually prevents drift.
- What does an evaluation gate look like for an agentic system in an enterprise? Competency questions and LLM-as-judge scoring get you a baseline. Turning that baseline into a release gate a business will respect is unsolved in most of the industry.
- How much of delivery management survives AI-assisted engineering? I now build as much as I coordinate. I do not yet know where that settles.
More detail on the Now page, which carries a review date.
Want to talk about a knowledge-system architecture? I am most useful early — when the question is still what to build and how it will be governed, not how to wire it.