Early-career researcher in LLM-based agents — closed-loop AI-for-AI training of mobile and web agents
2
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
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 scored: OpenAlex currently shows no aggregate output metrics to draw on beyond the 10 recent works. This axis would measure total publications, output cadence, and authorship roles across the full publication record.
Impact—
Not scored: citation-based measures (h-index 0, i10-index 0) reflect a record too new to interpret. This axis would measure citation accumulation, citing works, and influence within the LLM-agent research community.
Focus100/100
Focus scores 100/100: every OpenAlex topic on the record sits inside applied machine learning — Multimodal Machine Learning Applications (5), Advanced Neural Network Applications (4), Reinforcement Learning in Robotics (2), GAN-based image synthesis (2), and AI planning (2) — and the cataloged field distribution lists a single field, machine learning (LLM-based agents). The two catalog papers, on mobile and web agents, confirm one tightly unified research line.
Collaboration71/100
Collaboration scores 71/100, driven by breadth: 34 distinct coauthors appear in the recent window, and 12 recur often enough to be flagged as frequent — including Tingyu Qu, Weigao Sun, Yuecheng Liu, and Yucheng Zhao. On both cataloged papers Cao sits in the middle of the author list without corresponding status, a pattern typical of a contributing member of larger, multi-author agent teams rather than a solo research line.
Activity70/100
Activity scores 70/100: the recent three-year window contains 10 works, and counts-by-year data places all 10 in 2026 — a burst-style start rather than a steady multi-year cadence. Both cataloged papers also appeared on arXiv in 2026, indicating that essentially the entire visible record has been produced within a single highly active year.
Reproducibility20/100
Reproducibility scores 20/100, computed from a single assessed paper (mean 20). That paper, Qwen-Planner-Agent, carries a low reproducibility-readiness rating: Evolve's verdict notes that its headline MobilePA-Bench result of 77.05% is measured on a self-built benchmark and reported without significance tests, so independent verification of the claimed gains is currently difficult.
Research summary
Sihan Cao (Guangzhou Medical University, China; ORCID 0009-0005-2935-195X) is an early-career researcher whose OpenAlex-resolved record is recent and tightly concentrated: all 10 recorded works appear in 2026, with no citation accumulation yet (h-index 0, i10-index 0). The research line centers on large-language-model agents. Two 2026 arXiv papers sit in the Evolve catalog: Qwen-Planner-Agent, a closed-loop AI-for-AI framework that trains a 27B mobile planner to top MobilePA-Bench at 77.05%, and a budget-aware online-adaptation study for web agents. On both, Cao is a middle-listed, non-corresponding author, consistent with specialist contributions inside larger teams. OpenAlex topic assignments reinforce this orientation: Multimodal Machine Learning Applications (5 works), Advanced Neural Network Applications (4), plus Reinforcement Learning in Robotics, GAN-based image synthesis, and AI planning (2 each), while the cataloged field distribution points to LLM-based agents. Collaboration is unusually broad for this career stage — 34 distinct coauthors recently, 12 of them recurring. Evolve's review of the planner paper flags a reproducibility concern: headline gains rest on a self-built benchmark without significance tests, matching a low reproducibility score (mean 20/100 over 1 assessed paper). Output and impact dimensions are left unscored because the underlying data is insufficient; read this profile as a snapshot of an active, fast-forming research agenda rather than an established citation record.
Collaboration pattern
The collaboration pattern is team-based and wide: 34 distinct coauthors in the recent window, with 12 recurring partners — Tingyu Qu, Weigao Sun, Yuecheng Liu, Yucheng Zhao, Yi Zhu, Yifeng Ding, Qiyi Wang, Pengkun Jiao, Hanlei Xie, Xiongwei Wu, Qichao Wang, and Haodong Zhang. On both cataloged papers Cao contributes as a middle-listed, non-corresponding author, suggesting specialist roles within larger agent-research groups rather than a fixed two-person partnership.
Research keywords
Highlights
Closed-loop AI-for-AI training tops MobilePA-Bench
Reproducibility caution on a self-built benchmark
Budget-aware online adaptation for web agents
The second cataloged paper, When and What to Teach: Budget-Aware Online Adaptation for Web Agents (2026, arXiv), extends the agent agenda to web environments under an explicit budget constraint; it is not yet reviewed by Evolve, so field, keywords, and reproducibility remain unassessed.
Timeline
- 2026Qwen-Planner-Agent appears on arXiv — a closed-loop AI-for-AI framework training a 27B mobile planner that tops MobilePA-Bench at 77.05%.
- 2026A second arXiv paper, When and What to Teach: Budget-Aware Online Adaptation for Web Agents, extends the agenda to budget-constrained web agents.
- 2026OpenAlex records 10 works in this single year — the entire visible output to date — with 34 distinct coauthors, while citation counts remain at zero.
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.
- Bibliometrics come from OpenAlex only; h-index and i10-index are both 0, so the output and impact dimensions are not scored.
- Sihan Cao is a fairly common Chinese name; this profile is name-resolved from public OpenAlex bibliographic data, and the listed affiliation (Guangzhou Medical University) sits outside the ML-topics profile, so author disambiguation carries some uncertainty.
- Evolve review coverage is partial: only one of the two catalog papers has a verdict; the web-agents paper lacks field, keywords, and reproducibility assessment.
- The reproducibility score (20/100) rests on a single assessed paper, so it should be read as preliminary.