Tárcio André dos Santos Barros

Universidade Estadual de Campinas (UNICAMP) · Brazil

Unclaimed · AI-structured profileORCID OpenAlex

UNICAMP researcher (h-index 17) publishing high-reproducibility toolkits for LLM reliability and NLP interpretability

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 yet scored: this snapshot lacks the work-level publication data the output axis requires, so no output figure can be reported. Once linked, the axis would measure cumulative publication volume, publication venues and annual productivity against field baselines.

Impact—

Not yet scored: citation-based influence measures are unavailable in this record, so the impact axis is left unquantified. The axis would measure citations per work, field-normalized influence, and how often later studies build on this researcher's output.

Focus0/100

A focus score of 0 reflects strong dispersion: OpenAlex distributes the record across five near-equal topic clusters — electric motor design and analysis (47 works), photovoltaic system optimization techniques (40), sensorless control of electric motors (32), multilevel inverters and converters (30) and microgrid control and optimization (29) — with no dominant thread, and the two reviewed NLP toolkit papers fall outside these clusters, widening the spread.

Collaboration41/100

With 70 works over the last three years but only 7 distinct coauthors, the record shows low co-author density, and the collaboration score of 41 sits in the moderate range. Recurring partners include Frederico Jorge Heitmann, Daniel Omar Perez, José Luciano Verçosa Marques, Marcelo Vinicius de Paula and Reinaldo Cesar; on both reviewed papers the researcher appears as last author without a corresponding role, consistent with a senior supervisory pattern.

Activity100/100

The maximum activity score of 100 matches a steep trajectory in countsByYear: 1 work in 2012, 7 in 2015, 8 in 2018, 12 in 2023 and 13 in 2024, jumping to 40 works in 2025 with 17 already logged in 2026. With 70 works over the last three years, output is both sustained and accelerating, and the 2025 figure is the largest annual value in the record.

Reproducibility100/100

Reproducibility scores 100, with a reproMean of 100 across the 2 papers Evolve has reviewed. Both 2026 arXiv items are technical manuals rated high on reproducibility readiness, documenting toolkits step by step — one for confidence–corpus consistency via fine-tuning on a fabricated corpus, one for contextual individuation — though Evolve's verdicts caution that careful documentation does not substitute for reported empirical results.

Research summary

Tárcio André dos Santos Barros is listed at Universidade Estadual de Campinas (UNICAMP), Brazil, and this profile is name-resolved from public bibliographic data (OpenAlex) plus the papers Evolve has reviewed. The record shows an h-index of 17 and an i10-index of 26, with annual output climbing from 1 work in 2012 to 12 in 2023, 13 in 2024 and a peak of 40 in 2025, with 17 already recorded in 2026 and 70 works over the last three years. The record's dominant topics are electrical-engineering in character — electric motor design and analysis (47 works), photovoltaic system optimization techniques (40), sensorless control of electric motors (32), multilevel inverters and converters (30) and microgrid control and optimization (29) — while the Evolve-reviewed items sit in NLP: two 2026 arXiv technical manuals, one for a toolkit testing confidence–corpus consistency via fine-tuning on a fabricated corpus, the other measuring contextual word-sense individuation in transformer language models. Both were last-authored without a corresponding role and rate high on reproducibility readiness (reproMean 100 across 2 papers), though Evolve notes neither reports empirical results. The mismatch between bibliometric topics and reviewed NLP output is flagged as a possible record split; see caveats.

Collaboration pattern

The pattern is a small, stable circle: 70 works over the last three years against only 7 distinct coauthors, with Frederico Jorge Heitmann, Daniel Omar Perez, José Luciano Verçosa Marques, Marcelo Vinicius de Paula and Reinaldo Cesar recurring. On both reviewed papers the researcher signs as last author without a corresponding role — a senior, supervisory signature rather than a large distributed-collaboration style; the 41 score reflects this modest but steady team breadth.

Research keywords

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

Highlights

Toolkit for confidence–corpus consistency: methods manual via fine-tuning on a fabricated corpus
A 2026 arXiv technical manual for a toolkit testing whether LLM confidence tracks corpus consistency, using fine-tuning on fabricated arithmetic facts. Evolve calls it a careful methods manual but notes it reports no results; reproducibility readiness is rated high.
Measuring contextual individuation of word senses in transformer language models
A 2026 arXiv manual for a toolkit probing whether transformers individuate word senses by context through bridge forms, combining probing methodology with silhouette-coefficient analysis. Evolve judges the design well argued but lacking empirical validation.
Record output surge: 40 works in 2025
countsByYear jumps from 13 works in 2024 to 40 in 2025 — the largest annual figure in the record — with 17 more logged in 2026 to date and 70 works over the last three years.

Timeline

2012Earliest year in countsByYear, with 1 recorded work — the start of the visible publication record.
2015Output reaches 7 works, the first clear ramp-up after single-digit early years.
20188 works in 2018; the record then holds a steady rhythm of roughly 7–9 works per year through 2022.
2023Annual output rises to 12 works, then 13 in 2024, signalling an acceleration phase.
2025Peak year with 40 recorded works — the highest annual figure in the record.
202617 works already logged, including both Evolve-reviewed arXiv technical manuals on the confidence–corpus consistency and contextual individuation toolkits.

Papers in the Evolve catalog

Frequent co-authors

Frederico Jorge Heitmann×2 Daniel Omar Perez×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.

Bibliometric figures (h-index, i10-index, countsByYear, topics, coauthor counts) come from an OpenAlex record only; no citation-level impact data is available, which is why output and impact are unscored.
The OpenAlex topic profile is dominated by electrical-engineering topics (electric motors, photovoltaics, inverters, microgrids) while the reviewed papers and field assignment are NLP; the profile is name-resolved from public bibliographic data and this mismatch may indicate a split or conflated record.
Only 2 papers have been reviewed so far, both 2026 arXiv technical manuals that per Evolve verdicts report no empirical results; read this as an early-stage profile.
The 2026 figure (17 works) covers a partial year and will grow.