I'm Krystian Fernando, an applied AI research engineer.

Polinko's binary eval gates are determined by human judgement.

The gates generate signals that trace the pattern of its own behaviour as it's evaluated. The judgement stays attached to the evidence that produced the gate.

I created it because fluent answers can still invent sources, miss constraints, overstate certainty, or make direct work harder. Polinko preserves those failures as research material instead of smoothing them out of view.

I lead Polinko's research, engineering direction, and final claims.

I define the research questions, task boundaries, evidence criteria, and publication standard. I select what enters the public record, interpret the results, and decide what each claim can support.

This matters because Polinko is deliberately human-judged. The binary gates make judgement visible without handing authorship or responsibility to the systems being evaluated.

AI systems are collaborators, instruments, and part of what Polinko evaluates.

I use AI systems for dialogue, synthesis, reframing, bounded implementation support, validation, stress-testing, and documentation. Codex works with me inside the repository, while model APIs are part of the runtime evaluated across OCR, retrieval, response behaviour, and other bounded investigations.

The process is collaborative, but the judgement is mine. AI systems can help produce the work, and Polinko keeps their behaviour available for inspection at the same time.

The work connects research design, interaction design, implementation, and public explanation.

My broader practice looks at how people understand, evaluate, and collaborate with AI systems. Polinko is the current instrument for making those questions testable, documented, and visible.