Home on Artem Demchyshyn

Models are already smart enough for a huge amount of real work.

The AI boom keeps moving its own bottleneck: from compute, to memory, to interconnect, to power.

PDFBox should not be your architecture, and neither should Apache POI.

The twinned release: the same resolved layout now renders as a print-ready PDF and as an editable PowerPoint deck — native shapes, not a screenshot of a page — plus keep-with-next pagination and a render that can no longer destroy the file it replaces.

How fighting PDF coordinates turned into GraphCompose — a declarative, testable layout engine for structured PDFs in Java.

Convert a long Markdown document to PDF and it becomes something you move around in: a clickable bookmark outline, working [text](#heading) jumps, a page-numbered contents, bidirectional footnotes.

The illustrative release: native vector charts, SVG import with real gradients, free-form path clipping, and keepTogether() pagination — all compiled into the same deterministic engine primitives.

Four labs lead the AI race, but only two — OpenAI and Anthropic — have no ecosystem to fall back on.

Why document generation deserves to be treated as backend infrastructure, and when a Java project actually needs a document engine.

A short note on why I want a personal technical blog and what I plan to write about.

From fighting PDFBox coordinates to building a Java-first declarative document layout engine.

Engineering lessons from building layout, pagination, and rendering code in Java.

A practical view of AI-assisted development and where responsibility still belongs.

A short reflection on Claude Fable 5, Mythos-class models, safety layers, pricing, and what frontier AI access may become.