Yucheng Shen

—

Unclaimed · AI-structured profileORCID OpenAlex

Yucheng Shen — agentic visual RAG and LLM document generation; profile from two 2026 arXiv preprints

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—

Not scored: this snapshot holds no publication-count or productivity totals. The output axis would measure the volume and consistency of research output over time, and needs a fuller bibliographic record before any rating can be made.

Impact—

Not scored: no citation counts, h-index, or other influence metrics exist in the data. The impact axis would measure citation-based influence and uptake by the field; it cannot be inferred from two 2026 preprints.

Focus0/100

The focus axis scores 0 because the two reviewed papers split evenly between fields: the recorded distribution is 0.5 natural language processing (multimodal document generation) and 0.5 multimodal retrieval-augmented generation / vision-language agents. Both sit in adjacent LLM-agent territory, but two 2026 preprints are too thin a base to name one convergent specialty.

Collaboration53/100

Scoring 53/100, collaboration rests on 13 distinct coauthors across the two reviewed 2026 papers, with a visibly recurring core: Shuaiqiang Wang, Jianmin Wu, Lingyong Yan and Dawei Yin appear repeatedly in the frequency list, alongside Runze Li, Yukun Zhao, Can Xu, Jiulong Wu and Min Cao. A first-author position on SCORE suggests an active coordinating role, though no corresponding authorship is recorded.

Activity—

Not scored: the only dated records are two papers, both from 2026. The activity axis would measure publishing cadence and recent engagement over a multi-year window; a longer time series is required before any rating is possible.

Reproducibility38/100

The reproducibility mean is 38/100 over two scored papers (reproPapers: 2). PosterVisor is rated medium readiness, so its contract-and-rechecking pipeline is plausibly re-runnable; SCORE is rated low, since Evolve notes its three-benchmark state-of-the-art claim rests on an LLM-judge metric without variance estimates and partly non-comparable baselines. Verify artifact availability before any replication attempt.

Research summary

Yucheng Shen (name-resolved from public bibliographic data via ORCID 0000-0001-6539-4918; no institution or country on record) has an early-stage profile spanning two adjacent threads of large-language-model systems: multimodal document generation and multimodal retrieval-augmented generation with vision-language agents. The Evolve catalog holds two 2026 arXiv preprints. The first, SCORE (Navigating Sparse Evidence), is a first-author contribution that makes visual-evidence selection and consolidation an explicit, trainable stage of agentic visual RAG and reports state-of-the-art results on three benchmarks, with Evolve flagging an LLM-judge metric, missing variance estimates, and partly non-comparable baselines. The second, PosterVisor, a middle-author contribution, adds frozen rubric contracts and repair-aware rechecking to LLM poster generation; gains are judged real but modest, with ablation trade-offs. The field distribution across these papers is an even 50/50 split, which yields a focus score of 0 and signals breadth across two neighboring problems rather than a single established specialty. Collaboration data records 13 distinct coauthors and a recurring core (Shuaiqiang Wang, Jianmin Wu, Lingyong Yan, Dawei Yin), giving a moderate collaboration score of 53; no corresponding authorship is recorded. Reproducibility averages 38/100 across the two papers (one medium, one low readiness), so artifact availability is the main near-term uncertainty. Citation, output-volume, and activity metrics are not yet available.

Collaboration pattern

The collaboration pattern is a compact core plus a wider ring: 13 distinct coauthors across two 2026 papers, with Shuaiqiang Wang, Jianmin Wu, Lingyong Yan and Dawei Yin recurring most often in the frequency list, alongside Runze Li, Yukun Zhao, Can Xu, Jiulong Wu and Min Cao. Shen holds first authorship on SCORE and a middle position on PosterVisor; no corresponding authorship is recorded.

Research keywords

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

Highlights

First-author SCORE: agentic visual RAG with explicit evidence selection and consolidation
As first author, contributed SCORE, which turns visual-evidence selection and consolidation into an explicit, trainable stage of agentic visual RAG and reports state-of-the-art results on three benchmarks. Evolve's caveat: the gains rest on an LLM-judge metric with no variance estimates and partly non-comparable baselines, and reproducibility readiness is rated low.
Co-authored PosterVisor: contract-based control for LLM scientific poster generation
Contributed in a middle author position to PosterVisor, which adds frozen rubric contracts and repair-aware rechecking to LLM poster generation. Evolve judges the gains real but modest, with ablation trade-offs, and rates reproducibility readiness medium — the more re-runnable of the two reviewed systems.
A recurring pattern: making soft prompt-following explicit and checkable
Across both 2026 preprints, the work converts loose prompt-following into explicit, verifiable structures — frozen rubric contracts and recursive rechecking in poster generation, explicit selection and consolidation stages in visual RAG. Evaluation rigor (variance estimates, comparable baselines, released artifacts) remains the shared open gap flagged in both reviews.

Timeline

2026First-author preprint 'Navigating Sparse Evidence' (SCORE) appears on arXiv, proposing agentic visual RAG with explicit context selection and consolidation via GRPO reinforcement learning.
2026Co-authored preprint 'From Transient Prompts to Persistent Control' (PosterVisor) appears on arXiv, introducing recursive semantic-geometric contracts for scientific poster generation.
2026Both preprints enter the Evolve review catalog, with reproducibility readiness rated medium (PosterVisor) and low (SCORE), yielding a reproducibility mean of 38/100.

Papers in the Evolve catalog

Frequent co-authors

Shuaiqiang Wang×2 Jianmin Wu×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.

Bibliometric dimensions are computed from a limited snapshot of public bibliographic data; output, impact and activity axes have insufficient underlying data and are deliberately not scored.
The evidence base is two 2026 arXiv preprints only; no peer-reviewed venue, citation count, or h-index data is available.
Institution and country are not recorded; this profile is name-resolved from public bibliographic data via ORCID 0000-0001-6539-4918 and may not cover all work by people sharing this name.
Reproducibility signal covers only 2 papers (mean readiness 38/100) and reflects Evolve review judgments, not independent replication.