McGill researcher working on efficient LLM inference and self-evolving multi-agent systems for financial QA.
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-13) 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—
Output data is insufficient: beyond 11 recent works, no validated publication counts or venue-normalized productivity could be computed for this record. This axis would measure the volume and consistency of peer-reviewed output, and is deliberately left unscored here.
Impact—
Impact data is insufficient: the record shows h-index 0 and i10-index 0, and citation aggregates are not yet meaningful. This axis would measure citation uptake and influence on subsequent work, and is left unscored rather than estimated.
Focus0/100
The focus score is 0, reflecting a deliberately split portfolio: the field distribution is an even 50/50 between machine learning for efficient LLM inference and NLP/LLM agents for financial QA. Topic data reinforces the breadth — topic modeling (11), NLP techniques (9), information retrieval and search behavior (4), advanced neural network applications (3), and multimodal ML (2) — so work runs on two parallel threads rather than a single niche.
Collaboration73/100
Collaboration scores 73, backed by 38 distinct coauthors in the recent window. The frequent-coauthor list includes Charles Ling, Boyu Wang, Gang Niu, Qiuhao Zeng, Jerry Huang, and Yanzhang Ma, suggesting a dense, overlapping network. Both cataloged papers list Cui as a middle author rather than lead or corresponding, consistent with a contributor role inside larger teams; notably, Cui's own name also appears in the coauthor list, a sign of name-resolution noise.
Activity72/100
Activity scores 72. The snapshot records 11 works within the recent three-year window, with countsByYear showing all 11 dated 2026 — a burst of output concentrated in a single year. Five topic clusters are active over the same span, from topic modeling to multimodal ML. Because the observation window is short and no earlier record is present, this reads as a recently activated, high-tempo profile rather than a long-established one.
Reproducibility38/100
Reproducibility scores 38, the mean across the 2 papers Evolve has assessed (reproPapers: 2). ARM carries a medium readiness: efficiency gains look real, but baseline confounds weaken attribution. FINSKILLOPS is rated low: six-benchmark wins are tempered by author-built evals, judge bias, and unverifiable artifacts. The pattern suggests technically ambitious systems work whose evaluation protocols do not yet meet open-reproduction standards.
Research summary
Yufei Cui is a McGill University–affiliated researcher whose recent output sits at the intersection of efficient large language model inference and LLM-based agents. This profile is name-resolved from OpenAlex bibliographic data: the record shows 11 works in the recent three-year window, all dated 2026 in the snapshot, with topic mass concentrated in topic modeling (11), NLP techniques (9), and information retrieval and search behavior (4). The field distribution splits evenly (50/50) between machine learning for efficient LLM inference and NLP/LLM agents for financial QA. Evolve has reviewed two 2026 arXiv preprints on which Cui is a middle author and not corresponding: ARM, a differentiable routed-memory KV cache with gated writes and adaptive retrieval whose efficiency gains look real though baseline confounds weaken attribution; and FINSKILLOPS, a self-evolving multi-agent system for SEC filing QA whose six-benchmark wins are tempered by author-built evals, judge bias, and unverifiable artifacts. Bibliometric indicators are early-stage — h-index and i10-index both register 0 — while 38 distinct coauthors point to a heavily collaborative working style.
Collaboration pattern
Yufei Cui works in large, overlapping teams: 38 distinct coauthors appear in the recent window, with recurring partners including Charles Ling, Boyu Wang, Gang Niu, Qiuhao Zeng, Jerry Huang, and Yanzhang Ma. On both cataloged papers Cui is a middle author and not corresponding, suggesting a specialist-contributor role on systems projects rather than sole leadership. The reappearance of Cui's own name within the coauthor list indicates residual name ambiguity in the underlying data.
Research keywords
Highlights
Routed-memory KV cache for learnable sparse control in LLM inference
Self-evolving multi-agent system for SEC filing QA
Timeline
- 2026ARM preprint appears on arXiv, proposing a routed-memory KV cache with gated writes and adaptive retrieval for efficient long-context LLM inference.
- 2026FINSKILLOPS preprint appears on arXiv, presenting a self-evolving multi-agent system with gated skill maintenance for SEC filing QA.
- 2026OpenAlex records 11 works for this profile in 2026, the only year with activity in the snapshot.
- 2026Both reviewed papers enter the Evolve catalog, with reproducibility readiness rated medium for ARM and low for FINSKILLOPS.
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 derived from OpenAlex only; no ORCID is linked, so this profile is name-resolved from public bibliographic data and may conflate same-name authors.
- Citation-based impact cannot be assessed: the record shows h-index 0 and i10-index 0, and the output and impact radar axes are marked as missing.
- Only 2 of the 11 recent works carry Evolve reviews, so quality and reproducibility signals cover a minority of the output.
- Both reviewed papers are 2026 arXiv preprints with no peer-reviewed venue recorded, and 2026 is the only activity year in the snapshot, limiting trend inference.