Yuxin Zhang

—

Unclaimed · AI-structured profileOpenAlex

NLP and AI-agent-safety researcher working on calibrated small-model screening and agentic document-forgery benchmarks

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-09) 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—

Data insufficient: this axis measures publication volume and productivity over time. Only two Evolve-reviewed arXiv papers are on record here — no full publication list or counts — so no output score can be computed.

Impact—

Data insufficient: this axis measures citation-based influence, such as citation counts, h-index, and per-paper impact. No citation metrics are available in the current snapshot, so no impact score is reported.

Focus0/100

Scores 0/100, driven by an even field distribution: 50% natural language processing and 50% AI agent safety and document forensics across the two reviewed papers. With so few cataloged entries, this split reads as breadth across two adjacent areas — calibrated small-model NLP systems and agentic document-forgery evaluation — rather than an established specialization; additional papers would show whether the split is deliberate or transitional.

Collaboration48/100

Scores 48/100 with 10 distinct coauthors across the reviewed work. Both cataloged papers were written in teams with Yuxin Zhang in a middle, non-corresponding position, alongside repeat collaborators such as Simiao Ren, Ankit Raj, and Xingyu Shen. This points to steady participation in a moderately sized, stable group rather than solo authorship or a large cross-institutional network.

Activity—

Data insufficient: this axis measures recent research activity and publication cadence over time. The available records span only 2026, too narrow a window to estimate a meaningful activity trend or score.

Reproducibility55/100

Scores 55/100, the mean readiness Evolve assigned across two assessed papers (reproPapers = 2), each rated 'medium'. The scam-screening paper is candid about test-set exposure but rests entirely on synthetic data; the PDF-forgery study applies rigorous rule-based scoring yet is confounded by server-side heterogeneity. The profile is consistently mid-range: methods are documented and checkable, though neither release is fully turnkey.

Research summary

Yuxin Zhang is an early-stage researcher whose cataloged work splits evenly between natural language processing and AI agent safety with document forensics: a 50/50 field distribution across the two arXiv papers Evolve has reviewed, both from 2026, with Zhang in a middle, non-corresponding authorship position on each. The first, 'Open-Jev Judgments on CallScreenBench,' fine-tunes Qwen3-4B with LoRA and a JevLite typed-decision, label-token readout for calibrated one-pass scam-call screening; Evolve's review finds it non-inferior to an LLM judge at roughly 30x lower latency, while noting the evaluation is candid about test-set exposure but rests entirely on synthetic data. The second, 'Agents That Edit Documents,' measures how reliably LLM agents forge archived PDFs against a non-agentic control, reporting an 81.1% raw success rate that falls to 46.2% under strict filtering; its rule-based scoring is rigorous but confounded by server-side heterogeneity. Both papers carry 'medium' reproducibility readiness, averaging 55/100. Radar signals show 10 distinct coauthors (collaboration 48/100), a fully split field distribution (focus 0/100), and insufficient data for output, impact, and activity. Because 'Yuxin Zhang' is a common name and no ORCID, institution, or country is on record, this profile is name-resolved from public bibliographic data and should be read as an early-stage, partial view.

Collaboration pattern

Both cataloged papers were team efforts with Yuxin Zhang in a middle, non-corresponding authorship position. The work involves 10 distinct coauthors (collaboration score 48/100), including repeat collaborators Simiao Ren, Ankit Raj, and Xingyu Shen, plus Tommy Duong, Dennis Ng, Kidus Zewde, Yuchen Zhou, and Neo Tiangratanakul. The pattern suggests steady participation in a moderately sized, stable group rather than solo work or a large cross-institutional network; no institution or country is on record to confirm affiliations.

Research keywords

scam-call detectioncalibrationtyped decisionslabel-token readoutLoRA fine-tuningsmall language modelsagentic AI benchmarksPDF forgerydocument fraudrule-based verificationLLM agentscapability evaluation

Highlights

Small-model scam screening matches an LLM judge at ~30x lower latency
On CallScreenBench, a Qwen3-4B model fine-tuned with LoRA and a JevLite typed-decision label-token readout performs calibrated one-pass scam-call screening; Evolve's review finds it non-inferior to an LLM judge while running roughly 30 times faster, though all evaluation data is synthetic.
Quantifying agentic PDF forgery: 81.1% raw, 46.2% strict
The study measures how reliably LLM agents forge archived PDFs against a non-agentic control, reporting an 81.1% raw success rate that drops to 46.2% under strict filtering; the rule-based scoring design is rigorous, but server-side heterogeneity confounds ranking.
Both cataloged papers rated medium reproducibility (mean 55/100)
Evolve assessed reproducibility readiness for both 2026 papers and rated each 'medium', averaging 55/100 — a consistent mid-range profile in which methods are documented and checkable, though neither release is yet fully turnkey.

Timeline

2026Co-authored 'Open-Jev Judgments on CallScreenBench: Calibrated One-Pass Scam Screening with a Small Language Model' on arXiv (middle author, natural language processing).
2026Co-authored 'Agents That Edit Documents: Measuring Agentic PDF Forgery Against a Non-Agentic Control' on arXiv (middle author, AI agent safety and document forensics).
2026Both papers entered the Evolve review catalog, each rated 'medium' reproducibility readiness with a mean of 55/100.

Papers in the Evolve catalog

Frequent co-authors

Simiao Ren×2 Ankit Raj×2 Tommy Duong×2 Dennis Ng×2 Xingyu Shen×2 Kidus Zewde×2 Yuchen Zhou×2 Neo Tiangratanakul×2 Xingyu Shen×1 Ankit Raj×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 dimension scores and bibliometrics are computed from public bibliometric data only; citation and venue coverage may be incomplete.
Only the two arXiv papers reviewed by Evolve are covered; the output, impact, and activity axes lack sufficient data and are unscored.
'Yuxin Zhang' is a common name and no ORCID, institution, or country is on record — this profile is name-resolved from public bibliographic data and identity is not independently verified.
Per Evolve's review notes, the scam-screening evaluation relies entirely on synthetic data, and the PDF-forgery measurements are confounded by server-side heterogeneity.