Skip to content

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.

17+ years in enterprise data and delivery Teams of 8–70, distributed Databricks · AWS · Snowflake · Power BI 2nd place, EPAM Data Analytics Hackathon 2025

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:

DataGoverned ingestion, provenance, contracts at the boundary
SemanticsDefinitions with named owners, not a glossary nobody maintains
Retrieval & agentsRouting by intent, grounded answers, citations
DecisionsAnswers that land inside an operating workflow
Feedback & governanceEvaluation gates, regression on quality, ownership

Skip any stage and you get a demo. Close the loop and you get a system a business can put weight on.

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.

Read the case study →

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.

Read the pattern →

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.

Read the build note →

What I am working on now

Three questions I am actively testing, rather than a topic list:

  1. 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.
  2. 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.
  3. 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.

Get in touch → · Full CV → · Methods →