Early-stage NLP researcher in LLM and multimodal agent evaluation and data-centric agent fine-tuning
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-20) 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—
Insufficient data: the record contains only two 2026 arXiv preprints and no verified venue-level publication counts, so total scholarly output cannot be measured or scored. This axis would track cumulative papers, venues, and output volume across the researcher's career.
Impact—
No citation counts, h-index, or influence statistics appear in the available data, so research impact is not scored. This axis would measure citations, highly cited works, and scholarly uptake of the two cataloged LLM-agent papers.
Focus0/100
The focus score is 0 because the field distribution splits evenly at 0.5/0.5 between 'natural language processing / multimodal AI agent evaluation' and 'natural language processing (LLM agents)'. In practice both labels describe one LLM-agent research line — evaluating multimodal personal agents and refining multi-turn training trajectories — so true topical concentration is likely tighter than the score implies, and the dispersion is largely a field-labeling artifact.
Collaboration51/100
Collaboration scores 51, computed from 12 distinct coauthors in the bibliometric snapshot, a level consistent with moderate-sized team science typical of LLM-agent research. Both cataloged papers are middle-authored, non-corresponding contributions, and the frequent-coauthor list includes Yu Liu, Wenxiao Zhang, Cheng Hu, Cong Cao, and Kewei Tu, suggesting recurring but not yet dominant long-term partnerships.
Activity—
Only two papers, both dated 2026, are available, with no multi-year publication cadence, so recent activity cannot be scored. This axis would measure annual output rate, submission frequency, and research continuity across recent years.
Reproducibility55/100
Reproducibility scores 55, the mean across two papers assessed for reproducibility (reproPapers = 2), each rated medium readiness. Evolve's verdicts flag concrete gaps: MIRAGE's evidence rests on six artifacts and one runtime, bounding generality, while the trajectory-refinement paper's headline margins lack statistical support and its checkpoint-selection protocol favors the proposed methods.
Research summary
Xinyu Wang is an early-stage natural language processing researcher whose cataloged work centers on large language model (LLM) agents, spanning both evaluation and training-data engineering. Two 2026 arXiv papers are reviewed in the Evolve catalog, in both of which Wang is a middle author rather than corresponding author. The first, MIRAGE, is a controlled study of how conversation state shapes historical evidence use in multimodal personal agents, distinguishing pre- and post-compaction grounding-failure regimes. The second recasts multi-turn agent trajectory cleanup as deterministic editing of a round-level dependency DAG, reporting accuracy and efficiency gains for supervised fine-tuning. Bibliometric radar scores, computed from this small sample, place collaboration at 51 (from 12 distinct coauthors, with recurring partners including Yu Liu, Wenxiao Zhang, Cheng Hu, Cong Cao, and Kewei Tu), reproducibility at 55 (the mean across two papers, both rated medium readiness), and focus at 0, reflecting an even 50/50 split between two heavily overlapping NLP field labels. Output, impact, and activity axes lack sufficient data and are not scored. Given the commonality of the name, this profile is name-resolved from public bibliographic data and should be read as a provisional, early-stage record.
Collaboration pattern
Collaboration is moderate (radar 51, computed from 12 distinct coauthors). Both 2026 papers are middle-authored and non-corresponding, fitting a contributor role within larger LLM-agent teams. Recurring names include Yu Liu, Wenxiao Zhang, Cheng Hu, Cong Cao, Fangfang Yuan, Jin B. Hong, Yanbing Liu, Zhen Zhang, and Kewei Tu; the list also repeats Zhuo Chen, suggesting a frequent collaborator. Co-author roles and institutional ties are not documented in the available data.
Research keywords
Highlights
MIRAGE: conversation state as a driver of grounding failures in multimodal personal agents
Deterministic dependency-DAG editing for efficient multi-turn agent fine-tuning
Timeline
- 2026MIRAGE preprint posted to arXiv, a controlled study of how conversation state shapes historical evidence use in multimodal personal agents (middle author, non-corresponding).
- 2026Dependency-Aware Trajectory Refinement preprint posted to arXiv, recasting multi-turn agent trajectory cleanup as deterministic editing of a round-level dependency DAG (middle author, non-corresponding).
- 2026Both catalog entries fall in 2026, and output, impact, and activity metrics remain insufficient, marking this as an early-stage profile built on a two-paper record.
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.
- The name is common; this profile is name-resolved from public bibliographic data, and institution, country, and ORCID are not available in the input.
- All scores rest on only two reviewed papers, both 2026 arXiv preprints; no peer-reviewed venue record is present in the data.
- Output, impact, and activity radar axes are unscored because the underlying data is insufficient.
- The focus score of 0 reflects an even 50/50 split across two heavily overlapping field labels rather than genuinely scattered research topics.