ML researcher across LLM evaluation auditing, Bayesian network structure learning, and LLM-agent ecosystem studies.
4
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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 metrics are available yet: this profile lacks publication counts, venue distribution, and yearly totals, so the axis cannot be scored. It would measure the volume and consistency of the researcher's published output.
Impact—
Impact data such as citation counts, h-index, and downstream uptake are not present in the underlying bibliometric record, so this axis is left unscored. It would measure how far the published work influences follow-on research.
Focus0/100
The focus score of 0 reflects maximal breadth rather than absence of research: the field distribution splits evenly three ways — LLM-as-a-judge evaluation, probabilistic graphical models with LLM-augmented causal discovery, and LLM agents / empirical software engineering each hold exactly one third of the cataloged work. A unifying thread is the role of LLMs inside evaluation, causal modeling, and agent-maintenance pipelines.
Collaboration46/100
Scoring 46/100, collaboration sits mid-range: the bibliometric record links 9 distinct coauthors, with recurring names including Jackson Hassell, Farima Fatahi Bayat, Pouya Pezeshkpour, Chen Shen, EunJeong Hwang, Kushan Mitra, Dan Zhang, and Hannah Kim. All four cataloged papers place him as last author, none flagged corresponding, consistent with senior roles on small teams rather than large consortiums.
Activity—
Annual publication cadence cannot be computed: the catalog covers only 2026 preprints and no activity statistics exist in the source data. This axis would measure research pace and sustained productivity across years.
Reproducibility85/100
Reproducibility is the strongest recorded axis at 85/100, computed as the mean over 3 assessed papers: ABSOL and the agent-skill maintenance study rate high readiness, while MAWILE rates medium. Evolve's verdicts temper the number — ABSOL's largest-scale runs lack variance and MAWILE's evidence is broad but statistically thin — yet solid engineering and honest reporting recur across the portfolio.
Research summary
Estevam Hruschka (ORCID 0000-0003-1499-2808) is a machine-learning researcher whose recent work, resolved from public bibliographic data, spans three areas in near-equal measure: LLM-as-a-judge evaluation, probabilistic graphical models with LLM-augmented causal discovery, and empirical study of LLM-agent software ecosystems, each holding exactly one third of the cataloged output. Evolve's catalog holds four 2026 arXiv papers, all with him as last author. MAWILE builds a multi-axis workbench unifying judge-audit perturbations across prompt, rubric, input, and output; ABSOL applies bounded LLM guidance to scale Bayesian network structure learning to 1,041 nodes; a third study tracks who maintains public agent skill files; a fourth explores hypotheses-guided self distillation for continual personalization. Evolve's reproducibility assessment covers three papers and averages 85/100, with ABSOL and the skill-maintenance study rated high and MAWILE medium. The focus score of 0 reflects the evenly split three-way field distribution rather than any absence of research, and a mid-range collaboration score of 46 accompanies nine distinct coauthors. Citation-based output and impact measures are not yet available for this profile.
Collaboration pattern
The record shows 9 distinct coauthors, with Jackson Hassell, Farima Fatahi Bayat, Pouya Pezeshkpour, Chen Shen, EunJeong Hwang, Kushan Mitra, Dan Zhang, and Hannah Kim recurring most often. All four cataloged 2026 papers place Hruschka as last author, none marked corresponding, suggesting a senior-collaborator pattern on compact teams rather than large consortiums. The mid-range collaboration score of 46 is consistent with a small, stable set of partners spread across three distinct research lines.
Research keywords
Highlights
MAWILE: a multi-axis workbench for auditing LLM evaluators
ABSOL scales Bayesian network structure learning to 1,041 nodes with bounded LLM guidance
First longitudinal study of public agent skill-file maintenance
Timeline
- 2026Publishes MAWILE, a multi-axis workbench for inspecting LLM evaluators (last author, arXiv).
- 2026Publishes ABSOL, scaling Bayesian network structure learning to 1,041 nodes with bounded LLM orchestration.
- 2026Publishes the longitudinal study of human-governed, AI-assisted agent skill maintenance.
- 2026Publishes Hypotheses-Guided Self Distillation for Continual Personalization (not yet reviewed by Evolve).
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 dimensions are computed from public bibliometric data only; no citation counts, h-index, or venue-level impact figures are available in this snapshot.
- Institution, country, and career history were not provided in the source data and are therefore omitted here.
- The paper 'Hypotheses-Guided Self Distillation for Continual Personalization' has no field label, Evolve verdict, keywords, or reproducibility assessment yet.
- All four cataloged papers are 2026 arXiv preprints; peer-review status is not recorded, and the profile is name-resolved from public bibliographic data via ORCID 0000-0003-1499-2808.