Researcher in LLM-based multi-agent systems: graph topologies, agent memory co-evolution, and diagnostic benchmarking
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) 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 score yet: the snapshot contains only three 2026 arXiv preprints and no consolidated productivity totals, and this common name is resolved from OpenAlex alone. This axis would measure total works, venue record, and yearly output once the publication record stabilizes.
Impact—
Impact cannot be scored: no citation totals are provided and h-index and i10-index both read 0 in the snapshot. This axis would measure citation accumulation, landmark papers, and uptake by other groups once citation data accrues for the 2026 papers.
Focus8/100
The lowest axis at 8/100 signals extreme specialization. The field split is 66.7% LLM-based multi-agent systems and 33.3% their evaluation, and topic counts are led by Multi-Agent Systems and Negotiation (3) and AI-based Problem Solving and Planning (2), with Topic Modeling, NLP Techniques and Reinforcement Learning in Robotics at one each. Within the snapshot, Li works as a depth-first specialist on one problem family.
Collaboration54/100
Scoring 54/100, collaboration rests on 14 distinct co-authors in the snapshot. Minghao An, Zekai Chen, Rong-Hua Li and Kairui Yang recur most often, with Ziheng Yi, Kaixiang Zhang, Yuxuan Ba and Zhanke Liu also appearing. In all three reviewed 2026 papers Li sits in a middle author position without corresponding-author status, suggesting steady contribution within a compact recurring team rather than independent-led projects.
Activity40/100
At 40/100, activity reflects a very recent record: all three works in the snapshot fall in 2026, with 3 works in the recent three-year window and no earlier years captured. Preprints surfacing late in OpenAlex may understate ongoing work, but the observable record is a single-year burst of three arXiv papers rather than a sustained multi-year cadence.
Reproducibility55/100
Reproducibility readiness averages 55/100 across the three papers assessed, each rated medium. Evolve's verdicts temper the reported gains: MAGIC's first-place ranking on 8 benchmarks comes from single-run results, MACE's 81.11% versus SAGE's 78.97% rests on margins under 0.1 percentage points and uncited baselines, and OpenMAS-GCom's intervention evidence spans five of 29 datasets. Methods look promising; independent replication is not yet established.
Research summary
Xunkai Li is a researcher working on LLM-based multi-agent systems; this profile is name-resolved from OpenAlex bibliographic data and three 2026 arXiv papers reviewed by Evolve. The work is tightly concentrated on one problem family: MAGIC learns per-role atomic-or-group agent topologies through incremental construction with dense-reward reinforcement learning and ranks first on 8 benchmarks; MACE introduces memory-agent co-evolution over adaptive memory graphs, reporting 81.11% versus SAGE's 78.97%; and OpenMAS-GCom contributes a controlled-intervention diagnostic benchmark for graph-enhanced multi-agent systems spanning 29 datasets. The field distribution is correspondingly narrow — 66.7% LLM-based multi-agent systems and 33.3% their evaluation — producing a very low focus score (8/100). Activity (40/100) reflects three works, all recorded in 2026, the only publication year in the snapshot, with 3 works in the recent three-year window. Collaboration (54/100) spans 14 distinct co-authors, with Minghao An, Zekai Chen, Rong-Hua Li and Kairui Yang the most frequent; Li is a middle author without corresponding role on all three papers. Reproducibility readiness averages 55/100 (medium on all three), with Evolve flagging single-run evaluations and thin margins. Citation-based output and impact remain unscored because the snapshot shows h-index 0 and i10-index 0.
Collaboration pattern
Li publishes within a compact recurring cluster: 14 distinct co-authors overall, with Minghao An, Zekai Chen, Rong-Hua Li and Kairui Yang the most frequent partners, joined by Ziheng Yi, Kaixiang Zhang, Yuxuan Ba and Zhanke Liu. Across all three reviewed 2026 papers Li holds a middle author position without corresponding-author status, consistent with specialist contributions inside one team rather than leading independent groups.
Research keywords
Highlights
MAGIC: mixed-granularity agent graphs via incremental RL construction
MACE: memory-agent co-evolution with adaptive memory graphs
OpenMAS-GCom: diagnostic benchmark for graph-enhanced MAS
Timeline
- 2026MAGIC preprint appears on arXiv, introducing mixed-granularity agent graphs learned by incremental construction with dense-reward reinforcement learning (middle author).
- 2026MACE preprint appears on arXiv, presenting memory-agent co-evolution over adaptive memory graphs with reported 81.11% versus SAGE 78.97% (middle author).
- 2026OpenMAS-GCom preprint appears on arXiv, a controlled-intervention diagnostic benchmark for graph-enhanced multi-agent systems across 29 datasets (middle author).
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 are name-resolved from OpenAlex only; 'Xunkai Li' is a common name, so records may include name-ambiguity rather than verified identity.
- Citation-based output and impact metrics are unavailable (h-index 0, i10-index 0 in snapshot); those radar axes are left unscored.
- Institution, country and ORCID are missing from the source record, so affiliation context is not represented.
- Reviewed evidence covers three 2026 arXiv preprints; no peer-reviewed venue records appear in this snapshot.