Weihuang Zheng

—

Unclaimed · AI-structured profileOpenAlex

Early-career LLM-agent researcher: verifiable exploration benchmarks and evolving agent environments

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-09-28) 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—

Unscored: the snapshot gives recent3yWorks = 6 but lacks a verified full publication record and venue-quality data. This axis would measure cumulative output volume, first/corresponding-author share, and publication venues.

Impact—

Unscored: h-index and i10-index both read 0 in this OpenAlex snapshot, too thin for evaluation. This axis would measure citations, h-index trajectory, and how widely the reviewed benchmarks are adopted by other agent-evaluation work.

Focus—

Unscored: five OpenAlex topic labels exist (e.g., Scientific Computing and Data Management, Graph Theory and Algorithms), but the data is insufficient to quantify thematic concentration. This axis would measure how consistently output clusters around a core research line.

Collaboration66/100

Collaboration scores 66/100. The OpenAlex snapshot records 27 distinct coauthors over the recent three-year window, with 12 recurring collaborators including Ming Zhang, Shihan Dou, Junjie Ye, Maxm Pan, and Yuhui Wang. In both reviewed 2026 papers Weihuang Zheng is a middle author on large multi-author teams, indicating a group-based working style: steady contribution inside big LLM-agent projects rather than solo or corresponding-author leadership.

Activity57/100

Activity scores 57/100, grounded in 6 works over the recent three-year window and a counts-by-year series of 1 (2024), 1 (2025), and 4 (2026). The curve shows a clear ramp-up: 2026 is the most productive year so far, and both reviewed arXiv preprints — ExplorationBench and Breaking the Environment Wall — landed in that year.

Reproducibility—

Unscored: only two papers have been reviewed so far, rated low and medium for reproducibility readiness — too few to score. This axis would measure code and data release, artifact quality, and third-party replication.

Research summary

Weihuang Zheng is an early-career researcher working on large language model (LLM) agents, profiled here from OpenAlex bibliographic data. The snapshot shows 6 works in the recent three-year window with an accelerating output curve — 1 work in 2024, 1 in 2025, and 4 in 2026 — alongside 27 distinct coauthors, yielding a collaboration score of 66/100 and an activity score of 57/100. Two 2026 arXiv preprints reviewed in the Evolve catalog define the current focus: ExplorationBench, which measures AI systems' exploration in verifiable alien worlds with exact grading, and Breaking the Environment Wall (Env-Rethink), where a 27B verifier lifts noisy-environment agent pass rates by 13.3 percentage points across nine models and evolves harder tasks. Evolve's verdicts credit the benchmark design and system strength while flagging small-sample evaluation (n=3 trajectories, single-run controls) and an unproven recursive self-improvement claim. OpenAlex topic labels — Scientific Computing and Data Management, Graph Theory and Algorithms, plus Topic Modeling, Complex Network Analysis Techniques, and Business Process Modeling and Analysis — suggest a methods background that now concentrates on agent evaluation and environment engineering. Citation indicators (h-index and i10-index both 0) remain too thin to score impact.

Collaboration pattern

The collaboration profile is broad and team-based: 27 distinct coauthors in the recent three-year window (collaboration score 66/100), with 12 recurring collaborators including Ming Zhang, Shihan Dou, Junjie Ye, Maxm Pan, Yuhui Wang, and Zhangyue Yin. Both reviewed papers are large-team arXiv preprints where Weihuang Zheng appears as a middle author, consistent with contributing inside a sizeable LLM-agent research group rather than leading as first or corresponding author.

Research keywords

LLM agentsAgent environment designEnvironment evolutionAgent benchmarkingContext engineeringRecursive self-improvementVerifiable evaluationExploration in AI systemsFile verificationScientific computing and data managementComplex network analysisGraph theory and algorithms

Highlights

ExplorationBench: exact grading of AI exploration in alien worlds
A 2026 arXiv benchmark measures how AI systems explore in verifiable alien worlds with exact grading. Evolve credits the design as strong but notes n=3 trajectories and single-run controls limit statistical strength; reproducibility readiness is rated low.
Env-Rethink: a 27B verifier lifts noisy-environment agent pass rates by 13.3pp
In Breaking the Environment Wall, a 27B verifier lifts agent pass rates by 13.3 percentage points across nine models in noisy environments and evolves harder tasks, with reproducibility readiness rated medium; the recursive self-improvement claim remains unproven.
Methods-heavy topic profile now concentrates on LLM-agent work
OpenAlex topics pair Scientific Computing and Data Management (2) and Graph Theory and Algorithms (2) with Topic Modeling, Complex Network Analysis Techniques, and Business Process Modeling and Analysis (1 each); the two 2026 reviewed papers sit in LLM-agent evaluation and environment engineering, marking the current center of gravity.

Timeline

2024First visible output in the OpenAlex record: 1 indexed work.
2025A quiet year with 1 indexed work, consistent with an early-stage, still-accumulating record.
2026Output jumps to 4 works — the most productive year in the record so far.
2026Two arXiv preprints enter the Evolve catalog: ExplorationBench (LLM-agent exploration benchmarking) and Breaking the Environment Wall (evolving LLM-agent environments).

Papers in the Evolve catalog

Frequent co-authors

Maxm Pan×2 Ming Zhang×1 Zhenghao Xiang×1 Peizhong Gao×1 Yujiong Shen×1 Yuhui Wang×1 Zhonghan Yue×1 Shihan Dou×1 Zhangyue Yin×1 Junjie Ye×1 Shichun Liu×1 Jiahao Chen×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 indicators come from OpenAlex only; h-index and i10-index read 0 in this snapshot, which may reflect early-career status or name-disambiguation limits rather than actual influence.
No institution, country, or ORCID is available; the profile is name-resolved from public bibliographic data, and 'Weihuang Zheng' is a common name whose Chinese character form is unverified.
Only two reviewed papers (both 2026 arXiv preprints) are in the Evolve catalog, so highlights and reproducibility signals rest on a thin base.
Output, impact, focus, and reproducibility axes are unscored because the underlying data is insufficient.