Jianmin Wu

—

Unclaimed · AI-structured profileOpenAlex

Multimodal LLM researcher: contract-based poster generation and agentic visual RAG with explicit evidence selection.

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-09-20) 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—

Publication-count and venue data are not yet available for this profile; this axis would measure cumulative output — papers per year, venue tiers, and authorship roles — once a full publication record is reliably linked to this name.

Impact—

Citation counts and downstream-influence measures are absent from the underlying data, so no impact score is reported; this axis would track how often and how widely the two preprints are cited and built upon once a resolved citation record is available.

Focus0/100

The focus score is 0, the lowest concentration, and the underlying field distribution explains why: the cataloged work splits exactly evenly (0.5/0.5) between natural language processing (multimodal document generation) and multimodal retrieval-augmented generation / vision-language agents. Both papers are 2026 arXiv preprints in adjacent LLM-application areas, so with only two data points the profile shows breadth across two themes rather than a dominant specialization — a reading that may also reflect sparse, name-resolved data.

Collaboration53/100

The collaboration score is 53, built on 13 distinct coauthors in the resolved record. Recurring names include Shuaiqiang Wang, Lingyong Yan, and Dawei Yin (each appearing twice in the co-authorship list), alongside Yucheng Shen, Runze Li, Yukun Zhao, Can Xu, Jiulong Wu, and Min Cao. Wu sits in a middle authorship position on both papers without corresponding authorship, consistent with contributing inside a team rather than leading an independent group so far.

Activity—

Recent-activity signals such as publication cadence, submission streaks, and preprint-to-venue movement are not yet in the data; both cataloged papers carry 2026 dates, so this axis would measure output rhythm over multiple years once a longer record is linked.

Reproducibility38/100

The reproducibility score is 38, the mean over the 2 papers with reproducibility assessments. The spread matches Evolve's verdicts: PosterVisor's contract-and-recheck pipeline for poster generation is rated medium readiness — gains are real but modest with ablation trade-offs — while SCORE's agentic visual RAG results are rated low, given an LLM-judge metric without variance estimates and partly non-comparable baselines. Released code, data, and exact configurations would most move this score.

Research summary

Jianmin Wu is an early-documented researcher in multimodal large language model systems, profiled here from public bibliographic data on two arXiv preprints, both from 2026, on which Wu appears as a middle-listed author rather than first or corresponding author. The first, PosterVisor (From Transient Prompts to Persistent Control), tackles scientific poster generation, introducing frozen rubric contracts with repair-aware rechecking — recursive semantic-geometric contracts that persist beyond transient prompting; Evolve's review judges the gains real but modest, with ablation trade-offs, and rates reproducibility readiness medium. The second, SCORE (Navigating Sparse Evidence), makes context selection and consolidation an explicit, trainable stage of agentic visual retrieval-augmented generation, trained with GRPO reinforcement learning, and reports state-of-the-art results on three benchmarks; Evolve cautions that the case rests on an LLM-judge metric without variance estimates and partly non-comparable baselines, and rates reproducibility low. Across the catalog, Wu's work splits evenly (0.5/0.5) between multimodal document generation and multimodal RAG / vision-language agents, with 13 distinct coauthors — recurring names include Shuaiqiang Wang, Lingyong Yan, and Dawei Yin. Output, impact, and activity metrics are not yet available for this profile.

Collaboration pattern

Wu's resolved coauthorship network contains 13 distinct coauthors. The most recurrent names in the record are Shuaiqiang Wang, Lingyong Yan, and Dawei Yin, each listed twice, followed by Yucheng Shen, Runze Li, Yukun Zhao, Can Xu, Jiulong Wu, and Min Cao. On both cataloged 2026 papers Wu appears as a middle-listed author without corresponding authorship, indicating a contributor role within established teams rather than an independent lab lead so far. The collaboration score is 53.

Research keywords

scientific poster generationLLM agentscontract-based controlsemantic-geometric contractrecursive verificationlayout generationvisual retrieval-augmented generationagentic RAGevidence selectionevidence consolidationvision-language modelsGRPO reinforcement learning

Highlights

Explicit, trainable evidence selection for agentic visual RAG
SCORE turns context selection and consolidation into an explicit, trainable stage of visual retrieval-augmented generation, optimized with GRPO reinforcement learning, and reports state-of-the-art results on three benchmarks — though Evolve notes the case rests on an LLM-judge metric without variance estimates and partly non-comparable baselines.
Persistent contract control for LLM poster generation
PosterVisor replaces transient prompting with frozen rubric contracts and repair-aware rechecking — recursive semantic-geometric contracts — for scientific poster generation; Evolve judges the gains real but modest, with ablation trade-offs, and rates reproducibility readiness medium.

Timeline

2026Co-authored 'Navigating Sparse Evidence' (SCORE), making context selection and consolidation an explicit, trainable stage of agentic visual RAG, as a middle-listed author on arXiv.
2026Co-authored 'From Transient Prompts to Persistent Control' (PosterVisor), introducing frozen rubric contracts and repair-aware rechecking for scientific poster generation.
2026Cataloged work spans two evenly split fields — multimodal document generation and vision-language agents — within a 13-person coauthor network.

Papers in the Evolve catalog

Frequent co-authors

Yucheng Shen×2 Shuaiqiang Wang×2 Lingyong Yan×2 Dawei Yin×2 Runze Li×1 Yukun Zhao×1 Can Xu×1 Jiulong Wu×1 Min Cao×1 Lingyong Yan×1 Shuaiqiang Wang×1 Jianmin WU×1

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.

"Jianmin Wu" is a common name; this profile is name-resolved from public bibliographic data and may contain name-collision noise (the coauthor list itself includes a "Jianmin WU" variant).
Bibliometric scores derive from a sparse catalog of only two 2026 arXiv preprints; output, impact, and activity metrics are unavailable and unscored.
No institution, country, or ORCID is linked, so affiliation and identity verification remain pending.
The reproducibility score (38) averages only 2 assessed papers; Evolve verdicts note reliance on an LLM-judge metric without variance estimates on one of them.