University of Toronto researcher building process-based evaluation and shared-memory infrastructure for LLM agents
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
2
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-08) 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 scored: this axis measures publication volume and productivity trajectory across indexed works. The available OpenAlex record is too sparse for a reliable score; it would aggregate yearly work counts once more works accumulate.
Impact—
Not yet scored: impact tracks citations, h-index, and scholarly uptake of the work. The record shows an h-index of 2 and an i10-index of 0, too early-stage for meaningful scoring; no estimate is made.
Focus0/100
Scores 0/100 for topical concentration. The field distribution splits evenly — 0.5 agent evaluation and benchmarking, 0.5 LLM multi-agent systems and AI for research automation — and indexed OpenAlex topics scatter across materials science (6), chemical physics, drug discovery, quantum mechanics, and quantum computing (3 each), indicating breadth rather than a single concentrated line.
Collaboration61/100
Scores 61/100. The record lists 20 distinct coauthors, and recurring collaborators include Zhilin Wang, Shaokun Zhang, Yifan Zhang, Hao Zhang, Jin Xu, Binfeng Xu, Jian Hu, Karan Sapra, Andrew Tao, Jan Kautz, and Yi Dong. Both cataloged papers place the researcher in a middle author position, consistent with team-based work.
Activity79/100
Scores 79/100. OpenAlex counts 2 works in 2024, 2 in 2025, and 10 in 2026, with 14 works over the last three years — a clear 2026 acceleration that coincides with both cataloged agent papers, indicating a rapidly intensifying research pace.
Reproducibility55/100
Scores 55/100, averaged over the 2 papers Evolve has reviewed (reproMean 55), each rated medium readiness. OSWorld-Pro's judge validation is solid but rests on a single-judge, single-run design; Agora's 12-day, 13-agent run reaching 1.899 bpb has no causal test against a plain log — both temper replication confidence.
Research summary
Yunheng Zou is an early-career researcher at the University of Toronto whose work centers on the evaluation and infrastructure of LLM-driven agents. Two 2026 arXiv papers anchor the Evolve catalog: OSWorld-Pro, which introduces process-based evaluation for computer-use agents with 305 tasks, 2,814 subgoals, and 67k human labels; and Agora, which frames Git as shared memory for collective auto-research, demonstrated by a 12-day, 13-agent run reaching 1.899 bpb. Both entries position the researcher as a middle author, and Evolve rates both at medium reproducibility readiness (reproMean 55), noting that OSWorld-Pro's single-judge, single-run judging design and Agora's untested causal benefit over a plain log temper confidence. Bibliometric signals show a sharply accelerating pace: OpenAlex records 2 works in 2024, 2 in 2025, and 10 in 2026 (14 over the last three years), with 20 distinct coauthors including Zhilin Wang, Shaokun Zhang, Karan Sapra, Andrew Tao, Jan Kautz, and Yi Dong. The citation footprint is nascent (h-index 2, i10-index 0), so the output and impact axes remain unscored. OpenAlex topic clusters also surface materials-science and quantum topics that do not match the cataloged agent work, a likely author-disambiguation artifact. Overall, this is a fast-moving, team-based contributor to agent evaluation and multi-agent research automation.
Collaboration pattern
The profile lists 20 distinct coauthors, with a collaboration score of 61/100. Recurring collaborators include Zhilin Wang, Shaokun Zhang, Yifan Zhang, Hao Zhang, Jin Xu, Binfeng Xu, Jian Hu, Karan Sapra, Andrew Tao, Jan Kautz, and Yi Dong. On both cataloged 2026 papers the researcher sits in a middle author position with no corresponding role, consistent with contributing to larger team efforts rather than solo-led projects.
Research keywords
Highlights
Process-based evaluation at scale for computer-use agents
Git as shared memory for collective auto-research
Medium reproducibility readiness across both papers
Both cataloged papers carry a medium repro-readiness rating and an average reproducibility score of 55, reflecting substantial artifacts (human labels, agent logs) alongside single-run designs that limit independent replication confidence.
Timeline
- 20242 works recorded in OpenAlex — the earliest indexed output on this profile.
- 2025Indexed output held flat at 2 works; the citation footprint remained nascent (i10-index 0).
- 2026A sharp step-up: 10 indexed works in a single year, bringing the last-three-year total to 14.
- 2026Two agent papers enter the Evolve catalog — OSWorld-Pro and Agora — both with middle-author positions and medium reproducibility readiness.
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 drawn from OpenAlex only; identity is name-resolved from public bibliographic data (ORCID 0009-0007-9131-4468).
- Output and impact axes are unscored: an h-index of 2 and i10-index of 0 indicate a very early citation record.
- OpenAlex topic clusters (materials science, chemical physics, drug discovery, quantum topics) diverge from the cataloged agent/LLM work, suggesting possible author-disambiguation noise.
- Reproducibility scoring covers only the 2 papers Evolve has reviewed; broader venue peer-review status is not reflected.