Early-career contributor to open LLM post-training recipes and replayable forecasting-agent benchmarks
2
papers here
0
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-28) 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 snapshot contains no verified publication counts, venue tiers, or authorship shares. This axis would measure research output volume and cadence — papers per year, publication venues, and the share of first- or corresponding-author papers.
Impact—
Not yet scored: no citation counts, h-index, or adoption signals exist in this record. This axis would measure scholarly influence — citation trajectory over time, highly cited works, and uptake of released artifacts by other researchers.
Focus—
Not yet scored: the two reviewed papers point toward LLM post-training and forecasting agents, but the record lacks field-level distribution data. This axis would measure how concentrated or dispersed the publication portfolio is across topics and fields over time.
Collaboration67/100
The only scored axis, at 67/100, rests on 28 distinct coauthors in the underlying bibliometric record — a broad, team-heavy footprint. Twelve partners recur, including Yuan He, Chao Zhang, Liqin Ye, Haorui Wang, Fardin Ahmed, and Sudheer Chava. On both cataloged papers Zhang holds a middle author position and is not corresponding, consistent with specialist contributor roles inside larger groups rather than solo-led projects.
Activity—
Not yet scored: the only dated items in this snapshot are the two 2026 arXiv preprints, so yearly production rhythm cannot be computed. This axis would measure recent productivity and publishing momentum across the last several years.
Reproducibility—
Not yet scored: Evolve has not computed an aggregate reproducibility rating for this profile. The two reviewed papers do carry per-paper readiness grades — medium for Rufus-Air and high for Forecast-Dojo — which would feed this axis alongside code, data, and checkpoint releases.
Research summary
Rongzhi Zhang appears on Evolve as an early-stage, name-resolved profile: the catalog holds two 2026 arXiv preprints, both with Zhang in a middle author position, and no institution or country is resolved in the underlying record. The work centers on open, reusable infrastructure for large language models. Rufus-Air contributes an open eight-stage post-training recipe that Evolve's review credits with pushing GLM-4.5-Air-Base past the official GLM-4.5-Air release, while remaining unusually candid — though the rationale for the stage order is untested. Forecast-Dojo builds replayable environments for benchmarking and training LLM forecasting agents; Evolve deems its central finding credible and partly architectural, with research tools such as agent memory and belief updating helping all 12 evaluated models yet none surpassing the prediction market. Per-paper reproducibility readiness is rated medium and high respectively. Collaboration is the only scored radar dimension (67/100), grounded in 28 distinct coauthors and a recurring core of 12, including Yuan He, Chao Zhang, Liqin Ye, and Haorui Wang. Output, impact, focus, activity, and reproducibility remain unscored because citation counts, venue history, and year-by-year production data are absent from this snapshot. Readers should treat the profile as a bibliographic reconstruction that will firm up as more of the record is verified.
Collaboration pattern
With 28 distinct coauthors in the underlying record and a collaboration score of 67/100, Zhang works in sizeable, rotating teams. Twelve partners recur, including Yuan He, Chao Zhang, Liqin Ye, Haorui Wang, Fardin Ahmed, Ziyuan Lin, Yanbin Yin, Jing Peng, Michael Galarnyk, Sudheer Chava, Chia-Yuan Chang, and Renyuan Cheng. On both cataloged 2026 papers Zhang is a middle-position author rather than first or corresponding, suggesting a specialist contributor role within larger group efforts.
Research keywords
Highlights
Open post-training recipe surpasses the official GLM-4.5-Air release
Replayable forecasting benchmark: tools help all 12 models, none beats the market
A 2026 double release of reusable LLM infrastructure
Both cataloged outputs appeared as arXiv preprints in 2026 — one an open training recipe, the other a benchmark environment suite — indicating a reproducibility-first engineering style rather than single-model papers.
Timeline
- 2026Releases Forecast-Dojo, replayable environments for benchmarking and training LLM forecasting agents (arXiv preprint).
- 2026Releases Rufus-Air, an open eight-stage post-training recipe built on GLM-4.5-Air-Base (arXiv preprint).
- 2026Both dated outputs in the snapshot fall in this single year, and no earlier dated publications are recorded in the current record.
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 scores derive from OpenAlex-based public bibliographic data; no citation counts or h-index were available in this snapshot.
- No institution, country, or ORCID is resolved; 'Rongzhi Zhang' is name-resolved from public bibliographic data and may be ambiguous.
- Only the two Evolve-reviewed 2026 arXiv preprints inform this narrative; the wider publication record is not yet in the catalog.
- Five of six radar dimensions (output, impact, focus, activity, reproducibility) lack sufficient data and are unscored.