Gustavo Betarte

Universidad de la República de Uruguay

Unclaimed · AI-structured profileORCID OpenAlex

Machine-learning researcher at Universidad de la República spanning HTTP anomaly detection and LLM guardrail robustness

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-08) 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 publication-volume data is available for this axis. Output would measure cumulative and per-year paper counts across venues; with only two 2026 preprints in scope, no score is estimated here.

Impact—

Impact would measure citations, citing works, and downstream influence on the field. The current snapshot contains no citation or h-index data, so this axis is honestly left unscored rather than guessed.

Focus0/100

The focus axis scores 0, computed from the field distribution of reviewed work: 0.5 weight on AI security (LLM safety and adversarial robustness) and 0.5 on network security / machine learning (web application anomaly detection). With only two catalog papers, this even split reflects an exploratory two-track portfolio rather than a settled specialization; the score measures concentration of the current sample, not research quality.

Collaboration41/100

The collaboration axis scores 41, based on 7 distinct co-authors in the bibliometric snapshot. Recurring names include Fernando Outeda, Juan Diego Campo, Fiorella Cravero, Amanda Riverol, Rodrigo Martínez, and Álvaro Pardo. Both 2026 catalog papers place him in a middle author position without corresponding authorship, a pattern consistent with steady contribution inside a small, stable team rather than leadership of a large or international network.

Activity—

Activity would measure recent publication tempo and year-over-year research engagement. Only two 2026-dated preprints are on record, which is insufficient to characterize an ongoing cadence, so this axis remains unscored.

Reproducibility55/100

The reproducibility axis scores 55, the mean readiness rating across the 2 papers with reproducibility assessments, each individually rated medium. The HTTP embedding benchmark spans three datasets and multiple embedding models but was flagged for internal inconsistencies and no significance testing; the Prompt Guard 2 study is an explicitly exploratory, single-prompt case analysis. Both preprints look reusable in principle, yet evaluation hygiene limits confidence.

Research summary

Gustavo Betarte is a researcher at Universidad de la República de Uruguay whose reviewed work sits at the intersection of machine learning and security, organized around two empirical tracks. The first applies static word embeddings — FastText, Word2Vec, and Doc2Vec — to HTTP request anomaly detection: a 2026 arXiv preprint benchmarks these representations alongside one-class classifiers across three datasets, with Evolve's review crediting strong FastText consistency and coverage while flagging internal inconsistencies and missing significance testing. The second track probes the robustness of LLM guardrails: an XAI-guided perturbation case study of Prompt Guard 2, built on SHAP-style explanations, finds the detector relying on distributed lexical cues that synonym and paraphrase rewrites can flip, though its single-prompt scope limits generalization. In both catalog papers he appears in a middle author position, contributing within a compact team of seven distinct co-authors. Reproducibility averages 55 across the two reviewed papers, each rated medium readiness. This is an early-stage Evolve profile built from a small bibliographic snapshot; citation impact and longitudinal productivity are not yet assessed.

Collaboration pattern

Betarte publishes as a contributing middle author within a compact team: the snapshot records 7 distinct co-authors, with recurring collaborators including Fernando Outeda, Juan Diego Campo, Fiorella Cravero, Amanda Riverol, Rodrigo Martínez, and Álvaro Pardo. Neither 2026 catalog paper lists him as corresponding author, so current evidence points to a small, stable collaboration cluster rather than a large international network.

Research keywords

HTTP anomaly detectionstatic word embeddingsFastTextWord2VecDoc2Vecone-class classificationprompt injectionjailbreakLLM guardrailsPrompt Guard 2explainable AISHAP

Highlights

Benchmarking static embeddings for HTTP request anomaly detection
The 2026 arXiv preprint compares FastText, Word2Vec, and Doc2Vec representations with one-class detectors across three datasets. Evolve's review credits strong FastText consistency and coverage, while flagging internal inconsistencies and the absence of significance testing; reproducibility readiness is rated medium.
XAI-guided stress test of LLM prompt-injection guardrails
An exploratory explainability case study of Prompt Guard 2 using SHAP-based perturbation analysis shows the detector leaning on distributed lexical cues; synonym and paraphrase rewrites flip its verdicts. The single-prompt scope keeps the findings suggestive rather than conclusive.
A dual-track empirical ML-security agenda
Taken together, the two reviewed 2026 preprints pair web-facing anomaly detection with LLM guardrail robustness — both artifact-oriented evaluations of deployed ML defenses — sketching an early profile centered on empirically stress-testing security-critical machine learning.

Timeline

2026Co-authored an arXiv preprint benchmarking static embedding models (FastText, Word2Vec, Doc2Vec) for HTTP request anomaly detection, in a middle author position.
2026Co-authored an exploratory XAI-guided perturbation analysis of Prompt Guard 2 prompt injection detection, also as a middle author.
2026Both preprints entered the Evolve review catalog, pairing web-facing anomaly detection with LLM guardrail robustness in one early-stage profile.

Papers in the Evolve catalog

Frequent co-authors

Fernando Outeda×1 Gustavo Betarte×1 Juan Diego Campo×1 Fiorella Cravero×1 Amanda Riverol×1 Rodrigo Martínez×1 Álvaro Pardo×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 scores (collaboration 41, focus 0, reproducibility 55) derive from a limited public bibliometric snapshot; the output, impact, and activity axes have insufficient data and are left unscored.
The evidence base is the two 2026 arXiv preprints reviewed by Evolve; no citation counts, h-index, peer-reviewed venue record, or career history was provided.
Both catalog papers list this researcher in a middle author position without corresponding authorship; finer contribution details are unknown.
The frequent-coauthor list as supplied includes the profile subject's own name, treated as a data artifact; identity is name-resolved from public bibliographic data under ORCID 0000-0002-6863-1082.