Andre Bacellar

—

Unclaimed · AI-structured profileOpenAlex

IR researcher focused on cost-aware LLM reranking and failure prediction in multi-hop retrieval-augmented generation

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-13) 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 is computed. Available signals — six recent three-year works and a sparse counts-by-year record (one work in 2014, six in 2026) — are insufficient to score productivity. This axis would measure publication volume, consistency, and venue spread over time.

Impact—

No impact score is available. The i10-index is 0 and no citation aggregates beyond an h-index of 1 appear in the data, so influence cannot be scored. This axis would measure citations, uptake, and reference impact of the researcher's publications.

Focus0/100

The radar focus score is 0, the lowest value shown, and should be read against the raw distribution: fieldDist places all indexed field-assigned output in two adjacent retrieval-augmented-generation buckets (0.5 and 0.5), while topic assignments span five areas, led by topic modeling and information retrieval and search behavior (6 each), then recommender systems (3), multimodal machine learning (2), and construction project management (1). Interpret this axis with caution.

Collaboration14/100

Collaboration scores 14, the lowest computed axis. The OpenAlex record lists one distinct coauthor, and the frequent-coauthor list contains only the researcher's own name — an author-disambiguation artifact. Both catalog papers are first-author, corresponding-author works, so the reviewed output reads as effectively self-led with no identifiable repeat collaborator in the available data.

Activity57/100

Activity scores 57, mid-range. The counts-by-year record shows one work in 2014 and six in 2026, and the recent three-year window contains six works. The profile therefore reflects a long quiet stretch followed by a concentrated burst of 2026 output, including the two arXiv papers now in the Evolve catalog.

Reproducibility55/100

Reproducibility scores 55, computed from two papers (reproMean 55). Both Evolve-reviewed papers carry medium reproducibility readiness: the reranker-gating study reports a concrete fallback mechanism but exceeded its calibrated budget, while the confidence-scoring paper offers solid but small-margin results with thin theory — indicating moderately reusable artifacts.

Research summary

Andre Bacellar is an early-stage information retrieval researcher whose indexed work centers on retrieval-augmented generation. The OpenAlex-linked record carries an h-index of 1 and an i10-index of 0, with six works in the recent three-year window and an activity score of 57. The field distribution places all indexed field-assigned output in two adjacent retrieval-augmented-generation buckets (0.5 and 0.5), while topic assignments span five areas: topic modeling and information retrieval and search behavior (6 each), recommender systems (3), multimodal machine learning (2), and construction project management (1). The Evolve catalog holds two 2026 arXiv papers with Bacellar as first and corresponding author, both rated medium reproducibility readiness within a mean reproducibility score of 55 across two papers. The first proposes per-query gating of LLM rerankers for multi-hop retrieval, claiming roughly 51% fewer reranker calls; Evolve's review notes that only four of nine evaluation units met a 1pp non-inferiority threshold and the calibrated budget was exceeded. The second introduces score-distributional confidence scoring and abstention for multi-hop QA, judged solid and deployment-relevant though theoretically thin with small margins. Collaboration metrics are minimal — one distinct coauthor and a collaboration score of 14. No institution, country, or ORCID is provided, so this profile is name-resolved from public bibliographic data.

Collaboration pattern

Collaboration data is sparse: the OpenAlex record lists one distinct coauthor and the frequent-coauthor list contains only the researcher's own name, an author-disambiguation artifact. Both catalog papers are first-author, corresponding-author works, and the collaboration score is 14, the lowest computed axis. The reviewed output appears effectively self-led, with no identifiable repeat collaborator in the available data.

Research keywords

Retrieval-augmented generationMulti-hop retrievalLLM rerankingAdaptive RAGCost-aware routingGating classifierMulti-hop question answeringSelective predictionAbstentionConfidence calibrationANN score distributionTopic modeling

Highlights

Per-query gating claims ~51% fewer LLM reranker calls in multi-hop retrieval
The 2026 arXiv paper (first author, corresponding) proposes pre-call, per-query gating of LLM rerankers with an executable fallback mechanism, reporting savings of about 51% of reranker calls. Evolve's review credits the candid correction of early errors but notes only four of nine evaluation units met a 1pp non-inferiority threshold and the calibrated budget was exceeded; reproducibility readiness is medium.
Score-distributional confidence scoring for abstention in multi-hop QA
The second 2026 first-author, corresponding arXiv paper argues that cheap ANN-score distributional features can predictably flag multi-hop retrieval failures and support abstention. Evolve judges the results solid and deployment-relevant, while flagging thin theory and small margins; reproducibility readiness is medium.

Timeline

2014Earliest indexed work in the OpenAlex counts-by-year record, with one work listed for that year.
2026Two first-author, corresponding-author arXiv papers on multi-hop retrieval-augmented generation enter the Evolve catalog.
2026Six works are recorded in the recent three-year window; both catalog papers receive medium reproducibility readiness within a mean score of 55 across two papers.

Papers in the Evolve catalog

2026-09-19first authorcorresponding
Per-Query Gating of LLM Rerankers for Multi-Hop Retrieval
2026-09-18first authorcorresponding
Predictable Failure in Multi-Hop Retrieval: Score-Distributional Confidence Scoring and Abstention

Frequent co-authors

Andre Bacellar×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.

All bibliometrics come from a single OpenAlex snapshot; no ORCID, institution, or country is available, so this profile is name-resolved from public bibliographic data and may merge or split same-name authors.
Output and impact axes are unscored: the h-index of 1 and i10-index of 0 indicate a citation record too sparse to support those dimensions.
The frequent-coauthor list contains only the researcher's own name, an author-disambiguation artifact rather than evidence of collaboration.
The focus score of 0 sits in tension with the highly concentrated field distribution (two adjacent RAG buckets); interpret the focus axis with caution.