Alibaba Group researcher building scalable, verifiable environments for GUI, terminal, and hybrid LLM agents
3
papers here
2
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—
Not yet scored: this snapshot provides no publication counts, yearly production totals, or venue list for Que Shen. The output axis would measure cumulative and per-year paper volume, allowing comparison of research productivity against peers in the same field.
Impact—
Not yet scored: no citation counts, h-index, or citation history is available. The impact axis would measure how often publications are cited over time — a signal barely formed for 2026 arXiv preprints, which have had little time to accumulate citations.
Focus0/100
Focus scores 0, the lowest reading on this radar, consistent with the raw field distribution splitting evenly (0.5/0.5) between machine learning / LLM agents and machine learning for code / LLM agents. The catalog spans computer-use benchmark construction (RecreationWorld), terminal environment synthesis (Terminal-Universe), and LLM search-and-exploration methods — related but distinct problem areas, indicating breadth across the agent-environments agenda rather than one narrow specialty.
Collaboration80/100
The collaboration dimension scores 80, grounded in 54 distinct coauthors in the bibliometric record. All three catalog papers list Que Shen as a middle author within large teams, and eleven recurring collaborators appear, including Zhenru Zhang, Shuai Bai, Jiayong Deng, Sicheng Fan, Chang Gao, Mianqiu Huang, Yizhen Jiang, and Dehui Kong — a pattern typical of industrial lab groups building agent infrastructure at scale.
Activity—
Not yet scored: the snapshot contains three 2026 entries and no publication history. The activity axis would measure publication cadence over recent years — whether research engagement is accelerating, steady, or tapering — and needs a longer bibliographic record than is available.
Reproducibility60/100
Reproducibility scores 60, the mean across two assessed papers. The spread is wide: RecreationWorld is rated high for reproducibility readiness — a verifiable, five-platform benchmark environment — while Terminal-Universe rates low because its verifiers are teacher-authored and no code is released. Evolve also flags that RecreationWorld's transfer gains lack statistical controls, tempering the otherwise strong evaluation design.
Research summary
Que Shen is a researcher at Alibaba Group working at the intersection of machine learning and large language model (LLM) agents, with the profile's field distribution split evenly (0.5/0.5) between machine learning / LLM agents and machine learning for code / LLM agents. The Evolve catalog records three 2026 arXiv preprints, on all of which Que Shen is a middle-listed, non-corresponding author — a pattern consistent with contributing within a large industrial research team. Two reviewed works define a coherent agenda: building scalable, verifiable training and evaluation environments for agents. RecreationWorld introduces a five-platform recreation benchmark plus an SFT pipeline for hybrid GUI/coding agents, rated high for reproducibility readiness, though Evolve notes the transfer gains lack statistical controls. Terminal-Universe converts agent trajectories into scalable terminal environments, with ablations supporting +11.9 TB2.1 gains, but teacher-authored verifiers and no code release put its reproducibility readiness at low. The collaboration dimension scores 80, grounded in 54 distinct coauthors including recurring collaborators such as Zhenru Zhang, Shuai Bai, and Mianqiu Huang. Output, impact, and activity metrics are unavailable, so those radar axes remain unscored.
Collaboration pattern
Que Shen publishes within large Alibaba Group teams: all three 2026 catalog papers list a middle, non-corresponding author position. The bibliometric record shows 54 distinct coauthors, with a stable recurring core of eleven, including Zhenru Zhang, Shuai Bai, Jiayong Deng, Sicheng Fan, Yikun Fu, Chang Gao, Mianqiu Huang, Yizhen Jiang, Yuheng Jing, Dehui Kong, and Keliang Li. The collaboration dimension scores 80, indicating broad, team-based co-authorship around agent environments and training pipelines.
Research keywords
Highlights
Scalable, verifiable environments for hybrid computer-use agents
Turning agent trajectories into terminal training environments
Probing-guided search for diverse LLM solution exploration
A 2026 arXiv paper on answer probing-guided search for exploring diverse solutions from LLMs. Evolve has not yet published a review, keywords, or reproducibility assessment for this work, so it remains an unreviewed catalog entry.
Timeline
- 2026Co-authored RecreationWorld, a five-platform recreation benchmark and SFT pipeline for hybrid GUI/coding agents (middle author, arXiv preprint).
- 2026Co-authored Terminal-Universe, converting agent trajectories into scalable terminal training environments (middle author, arXiv preprint).
- 2026Co-authored Answer Probing-Guided Search for Diverse Solution Exploration of LLMs (middle author, arXiv preprint; 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 scores and the coauthor network derive from a public bibliographic snapshot only; no ORCID is provided, so this profile is name-resolved from public bibliographic data and could absorb homonymous authors.
- The output, impact, and activity radar axes are marked missing: no citation counts, h-index, or historical publication counts were provided, so those dimensions are left unscored.
- Only two of the three catalog papers carry Evolve review verdicts and reproducibility scores (reproPapers=2); Answer Probing-Guided Search is unreviewed, with no verdict, keywords, or slug.
- All catalog entries are 2026 arXiv preprints with middle, non-corresponding authorship; the profile is early-stage and long-term citation impact cannot yet be assessed.