Early-career researcher linking LLM agents, evaluation statistics, and software-engineering benchmarks at Northwestern
3
papers here
3
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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 scorable: the snapshot reports h-index 0 and i10-index 0 with only five works, all logged in 2026, so no reliable production estimate exists. This axis would measure publication volume, venue tier, and first- or corresponding-author share.
Impact—
Not yet scorable: a record that begins in 2026 shows h-index 0 and i10-index 0, so citation-based impact cannot be computed. This axis would measure citation accumulation and uptake of the benchmarks and bounds this work proposes.
Focus0/100
The focus score of 0 reflects maximal dispersion rather than absence of direction: the field distribution splits exactly evenly — 33.3% each for LLM-based agents, statistics for machine-learning evaluation, and software engineering / LLM evaluation — and the topic list spreads across five labeled topics, with Topic Modeling and Software Engineering Research at 3 mentions each, Natural Language Processing Techniques at 2, and two singletons. The through-line is evaluation rigor for LLM systems rather than any single subfield.
Collaboration46/100
A mid-range collaboration score of 46 rests on nine distinct co-authors within the recent window, including repeat collaborators such as Yezhou Cheng, Runjia Du, Zeming Liu, Hang Lyu, Yankai Zeng, Bojun Lin, and Yilan Wei. All three reviewed papers are multi-authored; the researcher holds first authorship on SemVerBench and middle positions on the other two, consistent with an active contributor in small teams rather than a solo or dominant author.
Activity52/100
Activity scores 52 on a thin but current base: recent3yWorks counts 5, and countsByYear places all five in 2026 with no earlier years recorded. The cadence — three arXiv preprints already reviewed by Evolve within that single year — indicates an active, burst-like start rather than an established multi-year output stream, so the moderate score should be read as an early-career baseline.
Reproducibility85/100
The strongest measured axis: reproducibility averages 85 across the three papers Evolve reviewed for repro readiness (reproMean 85, reproPapers 3). The hard-budget evaluation theory paper and SemVerBench are rated high readiness, while the agent-memory paper is medium. This fits a body of work built around oracle-labeled benchmarks and proved bounds, where artifacts and procedures are comparatively easy to re-run.
Research summary
Qibai Chen is an early-stage researcher at Northwestern University whose visible record, name-resolved from OpenAlex bibliographic data, begins in 2026: five works in the recent three-year window, all logged in 2026, with no citation accumulation yet (h-index 0, i10-index 0). Three arXiv preprints reviewed by Evolve span three equally weighted areas — LLM-based agents, statistics for machine-learning evaluation, and software engineering / LLM evaluation — each holding exactly one third of the field distribution. The work is cohesive despite the spread. The first-authored SemVerBench introduces a 240-item, oracle-labeled benchmark exposing systematic LLM blind spots in semantic-versioning constraint resolution; 'Sharp Limits for Honest Uncertainty in Hard-Budget Repeated Evaluation' proves sharp replication limits for confidence intervals under hard evaluation budgets; and 'Scope Before You Persist' addresses cross-family interference in persistent agent memory through scoped retrieval. Reproducibility is the strongest measured signal: a mean readiness of 85 across all three reviewed papers, two rated high and one medium. Collaboration is real but narrow — nine distinct co-authors in the recent window, with repeat collaborators including Yezhou Cheng, Runjia Du, Zeming Liu, Hang Lyu, Yankai Zeng, Bojun Lin, and Yilan Wei, and a mid-range collaboration score of 46. Authorship positions (first on one paper, middle on the others) are consistent with an early-career, team-embedded contributor.
Collaboration pattern
Nine distinct co-authors appear in the recent window, yielding a mid-range collaboration score of 46. The most frequent collaborators are Yezhou Cheng, Runjia Du, Zeming Liu, Hang Lyu, Yankai Zeng, Bojun Lin, and Yilan Wei, several of whom recur across records. All three reviewed papers are multi-authored arXiv preprints; Chen is first author on SemVerBench and a middle author on the other two, indicating an active small-team contributor rather than a solo or large-consortium author.
Research keywords
Highlights
First-authored 240-item benchmark exposes LLM blind spots in version-constraint semantics
Sharp replication limits proved for hard-budget LLM evaluation
Scoped retrieval proposed to stop cross-family interference in agent memory
Timeline
- 2026First-authored SemVerBench preprint appears on arXiv, benchmarking LLM comprehension of semantic-versioning constraint resolution on 240 oracle-labeled items.
- 2026Co-authored theory paper proves sharp replication limits for honest uncertainty in hard-budget repeated evaluation, rated high reproducibility readiness by Evolve.
- 2026Co-authored agent-memory paper proposes scoped retrieval to prevent cross-family interference in persistent memory, rated medium reproducibility readiness.
- 2026All five works in the recent three-year window fall within 2026, with no earlier output recorded in the per-year counts.
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 figures come from OpenAlex only; no ORCID is linked, so this profile is name-resolved from public bibliographic data.
- Output and impact dimensions are unscored: h-index and i10-index are both 0 and no citation counts are available in the snapshot.
- All three reviewed papers are 2026 arXiv preprints; no peer-reviewed venue information is present in the data.
- The record is thin (five recent works, three reviewed papers), making this an early-stage profile that will firm up as citations and venues accumulate.