Zhicheng Du

Tsinghua Shenzhen International Graduate School

Unclaimed · AI-structured profileOpenAlex

Applies LLMs to Traditional Chinese Medicine: prescription generation and multimodal diagnosis (Tsinghua SIGS)

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

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—

Output data are not yet available, so this axis is unscored. It would measure publication count, authorship roles, and venue spread over time; the current evidence base is limited to two 2026 arXiv preprints in the Evolve catalog.

Impact—

Impact data are not yet available, so this axis is unscored. It would measure citations and downstream uptake of the prescription generation pipeline and the LingLan diagnosis model; as 2026 preprints, citation signals have not accumulated yet.

Focus0/100

The focus score is recorded as 0. The underlying field distribution assigns 0.5 to 'medical AI / clinical NLP (TCM informatics)' and 0.5 to 'natural language processing (medical AI / TCM)' — adjacent labels describing the same TCM-plus-LLM agenda. With only two reviewed papers, the zero likely reflects splitting across near-synonymous categories rather than genuine topical breadth, so this axis should be read cautiously.

Collaboration36/100

The collaboration score is 36, computed from a compact footprint: 5 distinct coauthors across the cataloged papers. Recurring partners are Zheng Chen, Haoxuan Li, Peiwu Qin, and Yingshan Liang, and Du holds middle authorship on both papers without corresponding authorship. This pattern is consistent with a small, stable affiliated team rather than a broad cross-institution network.

Activity—

Activity data are insufficient for a score. This axis would measure publication cadence and year-over-year recency; with all reviewed work dated 2026, no multi-year productivity pattern can be established yet.

Reproducibility38/100

Reproducibility readiness averages 38 across 2 reviewed papers. The TCM prescription generation paper is rated medium: Evolve notes the pipeline is real but that its headline comparison rests on inconsistent PQS reporting and very narrow margins. LingLan is rated low, with malformed metrics and inconsistent confusion-matrix counts cited as barriers to independent verification.

Research summary

Zhicheng Du is an early-career researcher affiliated with Tsinghua Shenzhen International Graduate School whose cataloged work sits at the intersection of natural language processing and Traditional Chinese Medicine (TCM) informatics. The Evolve catalog contains two 2026 arXiv preprints, both listing Du in a middle authorship position without corresponding authorship. The first, 'Syndrome, Synergy, and Safety,' builds a structured-reasoning TCM prescription generation pipeline combining chain-of-thought distillation and DPO alignment with medication-safety constraints; Evolve's review finds the four-stage pipeline practically effective but cautions that its '7B surpasses GPT-5' headline rests on inconsistent PQS reporting and very narrow margins. The second, 'LingLan,' is a multimodal TCM diagnosis LLM trained with supervised fine-tuning and LoRA over tongue- and pulse-diagnosis data, reporting a 103.5% relative accuracy gain to 62.72% — figures Evolve flags for malformed metrics and inconsistent confusion-matrix counts. The recorded field distribution splits evenly between two adjacent medical-AI/NLP labels, and the collaboration footprint is compact: 5 distinct coauthors, with Zheng Chen, Haoxuan Li, Peiwu Qin, and Yingshan Liang recurring. Reproducibility readiness averages 38 across the two papers (medium and low). Output, impact, and activity metrics are not yet available, making this an early-stage, name-resolved profile.

Collaboration pattern

Both cataloged papers list Zhicheng Du in a middle authorship position without corresponding authorship, alongside a compact pool of 5 distinct coauthors. Recurring partners include Zheng Chen, Haoxuan Li, Peiwu Qin, and Yingshan Liang, suggesting a small, stable Tsinghua-affiliated team working across prescription generation and diagnosis modeling. The computed collaboration score is 36, consistent with a narrow but persistent co-author circle rather than a broad cross-institution network.

Research keywords

TCM prescription generationLarge language modelsStructured reasoningChain-of-thought distillationDPO alignmentMedication safetyMultimodal LLMSupervised fine-tuningLoRA fine-tuningTongue diagnosisPulse diagnosisTCM diagnosis

Highlights

Structured reasoning and knowledge-driven alignment for TCM prescription generation
A four-stage TCM prescription pipeline combining chain-of-thought distillation and DPO alignment with medication-safety constraints. Evolve's review finds it practically effective, but notes the reported '7B model surpassing GPT-5' headline depends on inconsistent PQS reporting and extremely narrow margins.
LingLan: a multimodal LLM for TCM diagnosis
A TCM diagnosis LLM fine-tuned with supervised fine-tuning and LoRA over multimodal tongue- and pulse-diagnosis data, claiming a 103.5% relative accuracy gain to 62.72%. Evolve flags malformed metrics and inconsistent confusion-matrix counts, and rates reproducibility readiness as low.
A consistent fine-tuning-and-alignment toolkit across both papers
The same methodological toolkit recurs across the catalog: chain-of-thought distillation, DPO alignment, supervised fine-tuning, and LoRA, applied to TCM generation and diagnosis tasks within one small coauthor team, with Du in a middle authorship position on both papers.

Timeline

2026Co-authored (middle position) 'Syndrome, Synergy, and Safety', a structured-reasoning TCM prescription generation pipeline posted to arXiv.
2026Co-authored (middle position) 'LingLan', a multimodal TCM diagnosis LLM built with supervised fine-tuning and LoRA over tongue- and pulse-diagnosis data.
2026Both papers entered the Evolve review catalog, receiving medium and low reproducibility-readiness ratings respectively.

Papers in the Evolve catalog

Frequent co-authors

Zheng Chen×2 Haoxuan Li×2 Peiwu Qin×2 Yingshan Liang×1 Haoxuan Li×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.

Radar scores are computed from a limited bibliometric snapshot; output, impact, and activity metrics were unavailable and are left unscored.
The reproducibility mean (38) rests on only 2 Evolve-reviewed papers, both 2026 arXiv preprints, so treat it as a small-sample signal.
No ORCID is linked and the profile is name-resolved from public bibliographic data; identity confirmation is pending.
Country is not provided in the source record, and affiliation details come only from the provided data.