Daniel Omar Perez

Universidade Estadual de Campinas (UNICAMP) · Brazil

Unclaimed · AI-structured profileORCID OpenAlex

UNICAMP-based researcher publishing high-reproducibility NLP toolkit manuals on LLM confidence and transformer polysemy

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-16) 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 scored: this OpenAlex record appears to mix entities — its counts-by-year stretch back to 1969 — so a clean per-year production count is unavailable. Once the record is disambiguated, this axis would measure annual output volume and its consistency over time.

Impact—

Not scored: citation evidence is too thin to rank — the snapshot reports an h-index of 4 and an i10-index of 1, likely diluted by name mixing, and the 2026 toolkits are too new for measurable uptake. This axis would measure citation influence and reuse of the toolkits.

Focus0/100

The focus axis scores 0/100, reflecting high dispersion in this record: OpenAlex topic mass spreads over at least five clusters — Kantian Philosophy and Modern Interpretations (32), Philosophy and Historical Thought (23), Philosophical Ethics and Theory (12), Epistemology, Ethics, and Metaphysics (11), and Psychoanalysis and Social Critique (9) — while the reviewed 2026 papers sit in NLP, with fieldDist split 0.5/0.5 between interpretability and LLM reliability.

Collaboration41/100

Scores 41/100. The last three years show 6 works with 7 distinct coauthors — a small-team pattern rather than large-consortium work. Recurring names in the record include Frederico Jorge Heitmann, Tárcio André dos Santos Barros, José Luciano Verçosa Marques, Marcelo Vinicius de Paula, and Reinaldo Cesar. Both catalogued papers are middle-authored and not corresponding, consistent with a contributing role.

Activity57/100

Scores 57/100 on sustained but uneven production: recent3yWorks is 6, and counts-by-year show 3 works in 2023, 1 in 2024, and 5 in 2026 — the last matching the two arXiv toolkit manuals. The long-tail record (a peak of 7 works in 2009, 5 in 2016) suggests intermittent bursts rather than continuous output.

Reproducibility100/100

Scores 100/100, the strongest axis: reproMean is 100 across 2 papers, and both reviewed manuals carry 'high' reproducibility readiness. This fits their genre — one specifies fine-tuning on a fabricated arithmetic-facts corpus, the other a silhouette-coefficient probing design — step-by-step manuals written so other teams can rebuild and run the toolkits, though neither yet reports empirical results.

Research summary

This profile is name-resolved from public bibliographic data (OpenAlex identity, ORCID 0000-0002-5965-3490) and covers Daniel Omar Perez of the Universidade Estadual de Campinas (UNICAMP), Brazil. The record has a dual character. Its OpenAlex topic mass sits in philosophy — Kantian Philosophy and Modern Interpretations (32), Philosophy and Historical Thought (23), Philosophical Ethics and Theory (12), Epistemology, Ethics, and Metaphysics (11), and Psychoanalysis and Social Critique (9) — while the works Evolve has reviewed are two 2026 arXiv technical manuals for NLP toolkits, and the computed field distribution splits evenly (0.5/0.5) between NLP interpretability and LLM reliability/hallucination. The snapshot reports an h-index of 4, an i10-index of 1, 6 works in the last three years, and 7 distinct coauthors; counts-by-year reach back to 1969, which suggests the record may blend name entities and should be read with caution. The reviewed manuals — one on fine-tuning against a fabricated arithmetic corpus to test whether LLM confidence tracks corpus consistency, one on probing contextual word-sense individuation in transformers via 'bridge forms' — are design-first documents with high reproducibility readiness (reproMean 100 over 2 papers) but no empirical results reported yet, making this an early-stage methods profile.

Collaboration pattern

The collaboration footprint is small-team: 6 works over the last three years with 7 distinct coauthors (collaboration score 41/100). Recurring names in the record include Frederico Jorge Heitmann, Tárcio André dos Santos Barros, José Luciano Verçosa Marques, Marcelo Vinicius de Paula, and Reinaldo Cesar. In both catalogued 2026 manuals the researcher is a middle author and not corresponding, consistent with contributing to a toolkit effort rather than leading the byline.

Research keywords

language model hallucinationconfidence calibrationfine-tuningsymbol groundingarithmetic reasoningfabricated corpuscontextual word representationspolysemyword sense individuationprobing methodologysilhouette coefficienttransformer language models

Highlights

Manual for testing whether LLM confidence tracks corpus consistency
Evolve reviewed this 2026 arXiv manual for a toolkit that fine-tunes on a fabricated corpus of arithmetic facts to test confidence–corpus consistency. The verdict: a careful methods document with high reproducibility readiness that reports no results — the empirical question remains open.
Probing contextual word-sense individuation in transformers
A second 2026 manual specifies a toolkit testing whether transformer language models individuate word senses by context using 'bridge forms', scored with silhouette coefficients. Evolve finds the design well argued but lacking empirical validation; reproducibility readiness is high.
Documentation-first, high-reproducibility publishing style
Both reviewed papers score high on reproducibility readiness, and the record's reproMean is 100 across 2 papers: the contribution is engineered as step-by-step manuals others can rebuild and re-run, ahead of any reported results.

Timeline

2009OpenAlex counts-by-year attribute a peak of 7 works to this name — the highest single year in the record.
2016A secondary burst of activity: 5 works are counted for this name in 2016.
20233 works are counted; the trailing three-year window totals 6 works with 7 distinct coauthors.
2026First reviewed output: an arXiv technical manual for a toolkit testing whether LLM confidence tracks corpus consistency via fine-tuning on a fabricated arithmetic corpus.
2026Second reviewed output: an arXiv manual for a toolkit measuring contextual word-sense individuation in transformer language models via 'bridge forms'.

Papers in the Evolve catalog

Frequent co-authors

Frederico Jorge Heitmann×2 Tárcio André dos Santos Barros×2 José Luciano Verçosa Marques×1 Marcelo Vinicius de Paula×1 José Luciano Verçosa Marques×1 Reinaldo Cesar×1 Marcelo Vinicius de Paula×1

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.

Bibliometrics come from the OpenAlex snapshot only; this profile is name-resolved from public bibliographic data and may blend entities sharing this name (counts-by-year reach back to 1969).
The record's dominant topics are philosophy (e.g., Kantian Philosophy and Modern Interpretations, 32), while Evolve-reviewed work is NLP; the two signals may reflect record mixing rather than a research shift.
Output and impact axes are unscored (dimMissing): citation indicators are thin (h-index 4, i10-index 1) and the 2026 toolkits predate measurable uptake.
Only two Evolve-reviewed papers exist (both 2026 arXiv manuals reporting no empirical results), so highlights and timeline underrepresent any unreviewed work.