Alibaba researcher building LLM agent environments and AI evaluation infrastructure; first author of Benchmark Radar
2
papers here
2
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 scored: this OpenAlex record has no reliable publication totals (h-index and i10-index are both 0). The axis would measure total works, venue mix, and per-year output once the profile is fully disambiguated.
Impact—
Not scored: no citation data is available for this profile, so impact cannot be estimated. The axis would measure total citations, citing works, and citation trajectory across the researcher's papers as the record accumulates.
Focus0/100
Focus scores 0, reflecting an unusually broad footprint: OpenAlex field distribution splits evenly (0.5/0.5) between AI-for-science agent environments and AI evaluation infrastructure, and the recorded topics span software engineering research, AI applications, topic modeling, and psychiatry/neuroscience. The two reviewed 2026 papers sit on opposite sides of this split — agent training environments versus benchmark search — so no single line yet dominates the record.
Collaboration79/100
Collaboration is the strongest recorded dimension at 79: OpenAlex credits 52 distinct coauthors, and 12 frequent collaborators are listed, including Hejia Geng, Zesen Huang, Haoyang Li, and Jiankai Sun. Both reviewed papers are large-team 2026 arXiv efforts — middle author on ScienceIDE and first author on Benchmark Radar — consistent with industry-scale group work anchored at Alibaba Group.
Activity57/100
Activity is mid-range at 57. The record shows 6 works in the recent three-year window, and countsByYear lists only 2026 with 6 publications, meaning all captured output is concentrated in a single year. Both Evolve-reviewed arXiv papers date to 2026, indicating a current burst of work, but the short window prevents any read on sustained cadence.
Reproducibility78/100
Reproducibility scores 78 (reproMean 78 across 2 assessed papers). Benchmark Radar is rated high readiness — a living database of 1,283 benchmark records with strong provenance — while ScienceIDE is rated medium, with real eval/SFT/RL gains but narrow coverage and selectively reported benchmark gains. Artifact availability and provenance discipline across both papers support the above-average score.
Research summary
Koutian Wu is a researcher at Alibaba Group (China) working at the intersection of LLM agent infrastructure and AI evaluation tooling. OpenAlex field distribution splits evenly (0.5/0.5) between machine learning / AI for science — agent environments and LLM training infrastructure — and AI evaluation infrastructure / information retrieval, with recorded topics spanning software engineering research, AI applications, topic modeling, and psychiatry/neuroscience. Two 2026 arXiv papers reviewed by Evolve anchor the profile: the first-authored Benchmark Radar, a living database and search engine over 1,283 AI benchmark records with a rare full-catalog census and strong provenance (though its own retrieval quality goes unevaluated), rated high for reproducibility readiness; and ScienceIDE, where Wu is a middle author, turning scientific codebases into agent-learnable environments with real eval/SFT/RL gains but narrow coverage and selectively reported benchmark gains, rated medium. Quantitatively, collaboration is the strongest recorded dimension (79), with 52 distinct coauthors and 12 frequent collaborators; reproducibility averages 78 across 2 assessed papers; activity is mid-range (57), with 6 works in the recent three-year window, all dated 2026. Output and impact are unscored: the record's h-index and i10-index are both 0, signaling a sparse profile rather than measured impact, and focus scores 0, reflecting the even two-field split.
Collaboration pattern
Collaboration is this profile's strongest recorded axis (79). OpenAlex credits 52 distinct coauthors, and the catalog lists 12 frequent collaborators, including Hejia Geng, Zesen Huang, Haoyang Li, Wenbin Li, Zihan Zhou, and Jiankai Sun. Both Evolve-reviewed papers are 2026 arXiv team efforts — Wu appears as middle author on ScienceIDE and first author (non-corresponding) on Benchmark Radar — a pattern of large, industry-affiliated group work with intermittent first-author leadership.
Research keywords
Highlights
Benchmark Radar: a living search engine over 1,283 AI benchmarks
ScienceIDE: turning scientific codebases into agent-learnable environments
Timeline
- 2026Co-authored ScienceIDE, converting the world's scientific codebase into agent-learnable environments with verifiable rewards and real eval/SFT/RL gains.
- 2026First-authored Benchmark Radar, a living database and search engine covering 1,283 AI benchmark records with a full-catalog census and strong provenance.
- 2026OpenAlex logged 6 works for this profile in 2026, with 52 distinct coauthors across the recent three-year window.
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; the record is sparse (h-index and i10-index both 0) and the profile is name-resolved from public bibliographic data, with no ORCID available to confirm identity.
- Output and impact dimensions are unscored because publication totals and citation data are missing from this record.
- Quality and reproducibility judgments rest on only 2 Evolve-reviewed papers, both 2026 arXiv preprints — an early-stage evidence base.
- countsByYear contains only 2026 (6 works), so long-term productivity trends cannot be assessed.