LLM evaluation methodologist building statistically grounded audits of generative search brand recommendations
4
papers here
2
reviewed by Evolve
4
first-author
3
corresponding
—
works · OpenAlex
3
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
Not yet scored: this axis would aggregate total publication volume, venue mix, and productivity trend over time. The current snapshot carries no computed output score, so no estimate is offered here.
Impact—
Not yet scored: this axis measures citation uptake such as h-index trajectory and per-paper influence. The available footprint (h-index 3, i10-index 1) is too thin to yield a stable score.
Focus0/100
Scores lowest among the computed axes (0/100). The field distribution splits evenly — 50% LLM evaluation methodology, 50% information retrieval / NLP evaluation — and indexed topics span five areas, led by topic modeling (17), information retrieval and search behavior (14), and ethics and social impacts of AI (12), then digital marketing (8) and NLP techniques (4). In practice the spread is narrower than the score implies, since both fields target LLM evaluation, but the multi-topic footprint supports the dispersion reading.
Collaboration14/100
Weak on this axis (14/100). The recent three-year record lists 32 works with only 1 distinct co-author, and all four catalogued papers are first-author efforts; Daniil Dzemesjuk is the one recurring collaborator. Corresponding-author status on three of the four papers confirms an essentially independent working model with minimal distributed teamwork.
Activity100/100
Maximum score (100/100). OpenAlex counts show 27 works indexed in 2026 alone — more than the 2008–2011 period and the preceding two years combined — following 2 works in 2024 and 3 in 2025, for 32 works across the recent three years. This is an intense, arXiv-driven burst concentrated on LLM auditing methodology.
Reproducibility55/100
Middle-band score (55/100) from 2 papers assessed, both rated medium reproducibility readiness. Evolve's review of the Dice Roll Method credits a statistically principled protocol with credible external validation but notes all data is the author's own and commercially tied; the finite-sample audit is called transparent and honest, yet verified only by AI without a human reference.
Research summary
Dmitrij Żatuchin is an Estonia-based researcher at the Estonian Entrepreneurship University of Applied Sciences whose work centers on measurement methodology for evaluating large language models in commercial settings. Writing as first author — corresponding author on three of four catalogued papers — he treats repeated LLM brand recommendations as a measurement-reliability problem rather than a one-off benchmark. His 2026 arXiv series includes the Dice Roll Method protocol (generalizability theory, negative-binomial GLMM), a finite-sample audit of brand and source discovery using rarefaction and Chao2 estimation, a study separating query language from exit-IP effects in a generative search interface, and a million-persona, intent-annotated buyer corpus for generative engine optimization. Bibliometric records show a sharp acceleration: after sparse indexed activity from 2008 to 2011 (1–2 works per year) and a multi-year gap, output rose to 2 works in 2024, 3 in 2025, and 27 in 2026 — 32 works over the recent three years. The citation footprint remains early-stage (h-index 3, i10-index 1). The work is essentially solo — one distinct recent co-author — and splits evenly between two closely related fields, LLM evaluation methodology and information retrieval, each at 50% of the field distribution.
Collaboration pattern
Dmitrij Żatuchin works almost entirely solo: the recent three-year window lists 32 works with a single distinct co-author, and the collaboration score is 14/100. Daniil Dzemesjuk is the one recurring collaborator in the record. All four catalogued 2026 papers are first-authored, three with corresponding-author status, indicating an independent protocol-author working model; no broader institutional or multi-team network is visible in the available data.
Research keywords
Highlights
The Dice Roll Method: standardized repeated-query auditing with generalizability theory
Finite-sample audit shows brand-discovery metrics depend on extraction choices
Query language and exit IP separated as factors in generative search recommendations
A 2026 first-author, corresponding-author study of a generative search interface showing that the language of the question and the exit IP act as separable factors shaping commercial recommendations — evidence that locale and phrasing jointly select the market a model displays. Not yet reviewed by Evolve.
A million-persona, intent-annotated buyer corpus for generative engine optimization
A 2026 first-author resource paper constructing and validating a million-persona, intent-annotated buyer corpus to support demand-side measurement for generative engine optimization, extending the auditing agenda from single repeated queries toward population-scale simulation. Not yet reviewed by Evolve.
Timeline
- 2008First indexed work appears in the OpenAlex snapshot (1 work); early output stays sporadic at 1–2 works per year through 2011.
- 2011Last indexed work before a multi-year publishing gap in the record (1 work this year).
- 2024Publishing resumes with 2 indexed works, ending the gap visible in the counts-by-year data.
- 2025Output grows to 3 indexed works, building toward the 2026 surge.
- 2026Peak year: 27 indexed works, including four first-author arXiv preprints — the Dice Roll Method protocol, the finite-sample brand-discovery audit, the query-language/exit-IP study, and the million-persona buyer corpus.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Bibliometric figures (h-index, counts by year, topics) come from an OpenAlex snapshot; no other citation source is included in this profile.
- Output and impact axes have no computed scores in this snapshot; the citation footprint (h-index 3, i10-index 1) is too small for stable estimation.
- Identity is name-resolved from public bibliographic data (OpenAlex, ORCID 0009-0002-1929-9770); Dmitrij Żatuchin may be a shared name.
- Two of the four catalogued papers await Evolve review (no verdict, field, or reproducibility data); the reproducibility score rests on only 2 assessed papers.