UNICAMP NLP researcher building reproducible toolkits for LLM confidence and contextual word-sense probing
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
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—
No output score yet: OpenAlex records h-index and i10-index as 0 and gives no aggregate productivity figure beyond five works in 2026. This axis would measure publication volume and cumulative research productivity over time; the underlying counts are not yet available.
Impact—
Impact is not yet scorable: no citation counts are available, and both cataloged manuals are 2026 arXiv preprints, too recent for uptake to accumulate. This axis would measure citation-based influence and scholarly diffusion of the toolkits.
Focus0/100
Focus scores 0 because the OpenAlex field distribution splits evenly, 0.5 to computational linguistics (NLP interpretability) and 0.5 to natural language processing (LLM reliability / hallucination). The two areas are closely adjacent, and topic counts — topic modeling (5), NLP techniques (5), adversarial robustness (3), explainable AI (2) — support a coherent NLP-centred agenda, but the exact even split drives the score to zero.
Collaboration41/100
Collaboration scores 41. Seven distinct coauthors appear in the recent window, with Daniel Omar Perez, Tárcio André dos Santos Barros, José Luciano Verçosa Marques, Marcelo Vinicius de Paula, and Reinaldo Cesar the most frequent partners. On both reviewed 2026 papers Heitmann sits in a middle author position and is not the corresponding author, indicating steady team participation rather than independent leadership so far.
Activity52/100
Activity scores 52. OpenAlex records five works in the trailing three-year window, and the counts-by-year data places all five in 2026 — a concentrated recent burst rather than a sustained multi-year record. Both Evolve-reviewed toolkits also carry 2026 dates, consistent with this early, compressed snapshot of output.
Reproducibility100/100
Reproducibility scores 100, the maximum: both cataloged papers carry high repro-readiness and the reproducibility mean across the 2 repro-scored papers is 100. As technical manuals they document toolkit design, procedures, and metrics in full — fabricated-corpus fine-tuning in one, silhouette-based probing in the other — though Evolve notes neither yet reports empirical results.
Research summary
Frederico Jorge Heitmann is a Brazil-based researcher at Universidade Estadual de Campinas (UNICAMP) working in natural language processing, with an OpenAlex field distribution split evenly (0.5/0.5) between computational linguistics (NLP interpretability) and NLP (LLM reliability / hallucination). The Evolve-reviewed catalog holds two 2026 arXiv technical manuals, each documenting a toolkit: one tests whether LLM confidence tracks corpus consistency by fine-tuning on fabricated arithmetic facts, and the other measures contextual word-sense individuation in transformer language models using 'bridge forms' and silhouette-based probing. Both are rated high repro-readiness, and the reproducibility mean across the 2 repro-scored papers is 100, although neither reports empirical results yet; Evolve notes the designs are carefully argued but unvalidated. Recorded activity is five works in the trailing three-year window, all dated 2026, with seven distinct coauthors and a middle, non-corresponding author position on both reviewed papers. Topic data points to topic modeling (5), NLP techniques (5), adversarial robustness (3), and explainable AI (2). Citation-based output and impact are not scored: h-index and i10-index are recorded as 0. This is best read as an early-stage, toolkit-building profile with strong documentation practice and no citation track record yet.
Collaboration pattern
Heitmann works in small teams: seven distinct coauthors appear in the recent OpenAlex window, with Daniel Omar Perez, Tárcio André dos Santos Barros, José Luciano Verçosa Marques, Marcelo Vinicius de Paula, and Reinaldo Cesar as the most frequent partners (Marques and de Paula each recur twice in the frequent-coauthor list). On both Evolve-reviewed 2026 manuals he holds a middle, non-corresponding author position, consistent with a contributor role in a UNICAMP-centered NLP group.
Research keywords
Highlights
Toolkit manual for LLM confidence–corpus consistency
Measuring contextual word-sense individuation in transformers
Timeline
- 2026Published a technical manual for a toolkit measuring contextual word-sense individuation in transformer language models via 'bridge forms' (arXiv).
- 2026Published a technical manual for a toolkit testing LLM confidence–corpus consistency by fine-tuning on a fabricated arithmetic corpus (arXiv).
- 2026OpenAlex records five works dated 2026 within the trailing three-year window, with seven distinct coauthors — the entirety of the visible recent output.
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 come from OpenAlex only (identity: openalex), a source that may omit or merge works.
- h-index and i10-index are recorded as 0 with no citation counts available, so the output and impact axes are left unscored.
- Only two 2026 papers are Evolve-reviewed, both technical manuals that report no empirical results; broader catalog coverage may change this picture.
- The profile is name-resolved from public bibliographic data (ORCID 0009-0006-5828-8301) and contains no verified biographical details.