Researcher working across NLP and AI agent safety: small-model scam screening and agentic PDF-forgery measurement
2
papers here
2
reviewed by Evolve
0
first-author
0
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—
No publication-volume metrics are available: total paper count, venue types, and yearly output are absent from this snapshot, so the output axis is unscored. It would measure cumulative publication productivity and its distribution over time.
Impact—
No citation counts, h-index, or download statistics exist in the input, so influence cannot be scored. The impact axis would measure citations, uptake, and downstream reuse of this researcher's work by the wider community.
Focus0/100
The radar focus score is 0, consistent with the raw field distribution: work splits evenly (0.5/0.5) between AI agent safety and document forensics and natural language processing. Both 2026 papers sit in these distinct threads — one on scam-call screening with small language models, the other on agentic PDF forgery — so output is spread across topics rather than concentrated in a single line of research.
Collaboration48/100
Collaboration scores 48, a mid-range value built on 10 distinct coauthors in the bibliometric record. The frequent-coauthor list names Simiao Ren, Ankit Raj, Tommy Duong, Yuxin Zhang, Dennis Ng, Kidus Zewde, Yuchen Zhou, and Neo Tiangratanakul. On both reviewed 2026 papers Shen holds a middle authorship position and is not corresponding author, indicating steady team-based work without yet a clear leading role.
Activity—
Year-by-year publication counts are not provided, so recent research cadence cannot be scored. The activity axis would measure publishing frequency and continuity over recent years; only the two 2026 arXiv entries are dated in this snapshot.
Reproducibility55/100
Reproducibility scores 55, the mean over 2 scored papers, both rated medium readiness by Evolve review. The scam-screening paper is candid about test-set exposure but relies entirely on synthetic data; the PDF-forgery study uses a rigorous rule-based scoring design, though server-side heterogeneity complicates replication and ranking. Together this suggests reusable methodology with known, documented limitations.
Research summary
Xingyu Shen is an early-stage researcher whose reviewed work to date sits at the intersection of natural language processing and AI agent safety, split evenly (0.5/0.5) between the two fields in the bibliometric field distribution. Two 2026 arXiv preprints are in the Evolve catalog, on neither of which Shen is first or corresponding author — both list a middle authorship position. The first applies JevLite typed-decision readout to a LoRA fine-tuned Qwen3-4B for one-pass scam-call screening, reported as non-inferior to an LLM judge at roughly 30x lower latency; the Evolve review notes the paper is candid about test-set exposure but that all evaluation data is synthetic. The second measures how reliably LLM agents can forge archived PDFs against a non-agentic control, reporting an 81.1% raw success rate that falls to 46.2% under strict filtering; reviewers credit a rigorous rule-based scoring design while flagging server-side heterogeneity as a confound for rankings. Radar scores place collaboration at 48 (10 distinct coauthors), reproducibility at 55 across 2 scored papers (both rated medium readiness), and focus at 0, reflecting the balanced two-field spread. Output, impact, and activity axes lack sufficient data and are left unscored. Because no ORCID, institution, or country is available, this profile is name-resolved from public bibliographic data.
Collaboration pattern
Collaboration is team-based: the bibliometric record lists 10 distinct coauthors, and the frequent-collaborator set includes Simiao Ren, Ankit Raj, Tommy Duong, Yuxin Zhang, Dennis Ng, Kidus Zewde, Yuchen Zhou, and Neo Tiangratanakul. The radar collaboration score is 48, mid-range. On both reviewed 2026 arXiv papers Shen appears in a middle authorship position rather than first or corresponding author, consistent with contributing within a larger group rather than leading it.
Research keywords
Highlights
Non-inferior scam screening at ~30x lower latency
Quantifying agentic PDF forgery against a non-agentic control
Rule-based scoring as a reproducible yardstick
Timeline
- 2026Co-authored 'Open-Jev Judgments on CallScreenBench: Calibrated One-Pass Scam Screening with a Small Language Model' (arXiv) in a middle authorship position; Evolve rates its reproducibility readiness medium.
- 2026Co-authored 'Agents That Edit Documents: Measuring Agentic PDF Forgery Against a Non-Agentic Control' (arXiv), reporting an 81.1% raw and 46.2% strict forgery success rate; also middle authorship, medium readiness.
- 2026Within this single documented year, output spans two fields evenly (0.5/0.5) — NLP and AI agent safety/document forensics — with 10 distinct coauthors on record and no first/corresponding authorship yet.
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.
- Name is common and no ORCID, institution, or country is available; this profile is name-resolved from public bibliographic data and may mix records of different people.
- Radar scores are computed from public bibliometric data; the underlying source index is not specified in this snapshot.
- Only two 2026 arXiv preprints are reviewed here; output, impact, and activity axes are unscored because no citation or yearly-volume data was provided.
- The supplied frequent-coauthor list contains the researcher's own name and a duplicate entry (Ankit Raj twice), indicating coauthor-disambiguation noise.