Marian Alexa

Working on the right questions.

I get paid to find the thing that's wrong.

Model outputs, media, arguments — noticing what doesn't fit.

Currently

I evaluate AI systems for the ways they fail: synthetic voice and video passing as real, reasoning that holds together right up until it doesn't, answers that are fluent and wrong, outputs that should never have shipped. The domain changes from one week to the next — media authenticity, a technical review, a rubric nobody has stress-tested yet. The habit doesn't: refusing to take a thing part by part, and checking whether the parts add up to the whole they claim.

Work

Synthetic media

Voice deepfake detection; image and video authenticity.

Reasoning

Multi-step verification — factuality, and the joins where a chain quietly stops holding.

Safety

Unsafe and harmful output review against rubrics that change under you.

Code

Correctness and logic review, at volume.

Judgment at scale

Structured label frameworks, edge cases, and staying consistent across thousands of items.

Now

RWS

AI Analyst — synthetic media detection, reasoning verification, output safety review.

Alignerr

AI Analyst, Subject Matter Expert & Code Reviewer.

Both roles are vendor-side, on consumer AI products.

Before

Relay Cars

Operations Manager, ten years. London.

London Met

BA (Hons) Banking & Financial Economics.

Reach

1ai@pm.me LinkedIn ↗