NLP and AI agent safety researcher: calibrated small-model scam screening and benchmarks for agentic PDF forgery
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—
Not yet scored: the snapshot holds only the two reviewed arXiv papers, with no total publication count, venue history, or per-year output volume. This axis would measure the volume and consistency of research output once a complete publication record is linked.
Impact—
Not yet scored: no citation counts, h-index, or per-paper citation data are present in this snapshot, so influence cannot be estimated. This axis would measure how widely the work is cited and reused once citation records are linked.
Focus0/100
Scores 0 on field concentration: the 50/50 field distribution splits evenly between AI agent safety and document forensics and NLP. Rather than deep single-field specialization, the record so far covers two adjacent LLM-facing areas — calibrated small-model screening on one side, agentic document-forgery benchmarking on the other — a broad early agenda that may consolidate as more papers accumulate.
Collaboration48/100
Scores 48, computed from 10 distinct coauthors in the snapshot. The frequent-coauthor list is compact — Ankit Raj and Xingyu Shen each recur twice, alongside Simiao Ren, Tommy Duong, Yuxin Zhang, Dennis Ng, Kidus Zewde, and Neo Tiangratanakul. Both catalog papers place this researcher in a middle, non-corresponding position, consistent with steady contribution within a recurring collaboration group rather than sole leadership so far.
Activity—
Not yet scored: the snapshot has no publication dates beyond the two 2026 arXiv entries and no multi-year productivity series. This axis would measure research cadence — papers per year and continuity of output — once a longer publication timeline is linked.
Reproducibility55/100
Scores 55, the mean readiness across the 2 papers carrying a reproducibility rating; each is rated medium. The scam-screening paper is candid about test-set exposure but all its data is synthetic, while the PDF-forgery paper's rule-based scoring is rigorous though server-side heterogeneity confounds ranking. Releasing code, data, or scoring harnesses would raise this axis.
Research summary
Yuchen Zhou works across natural language processing (NLP) and AI agent safety and document forensics, with the visible record anchored to two 2026 arXiv papers reviewed in the Evolve catalog. The first, Open-Jev Judgments on CallScreenBench, runs a JevLite typed-decision readout — trained with LoRA fine-tuning — on Qwen3-4B 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 candidly that the evaluation data is entirely synthetic. The second, Agents That Edit Documents, measures how reliably LLM agents forge archived PDFs against a non-agentic control: an 81.1% raw success rate falls to 46.2% under strict filtering, and its rule-based scoring design is judged rigorous even though server-side heterogeneity confounds rankings. The field distribution splits evenly, 50% NLP and 50% agent safety and document forensics, forming a thematically coherent pair around calibrated, verifiable LLM behavior. The snapshot lists 10 distinct coauthors, with Ankit Raj and Xingyu Shen recurring most often; both papers list Zhou in a middle, non-corresponding author position. Reproducibility averages 55 across two medium-readiness ratings, and the output, impact, and activity axes remain unscored for lack of data. 'Yuchen Zhou' is a common name, so this profile is name-resolved from public bibliographic data.
Collaboration pattern
The collaboration footprint scores 48 from 10 distinct coauthors. Partnerships are compact and recurring: Ankit Raj and Xingyu Shen each appear twice in the frequent-coauthor list, alongside Simiao Ren, Tommy Duong, Yuxin Zhang, Dennis Ng, Kidus Zewde, and Neo Tiangratanakul. Both catalog papers place Zhou in a middle, non-corresponding author slot, so the visible pattern is steady contribution within a small recurring team rather than independent or corresponding-author leadership.
Research keywords
Highlights
Non-inferior scam screening at ~30x lower latency
Quantifying how reliably LLM agents forge archived PDFs
Timeline
- 2026Co-authored the CallScreenBench scam-screening paper (middle author, arXiv), pairing calibrated typed-decision readouts with a small language model.
- 2026Co-authored the agentic PDF-forgery benchmark paper (middle author, arXiv), documenting 81.1% raw and 46.2% strictly filtered forgery success rates.
- 2026Combined record shows an even 50/50 footprint across NLP and AI agent safety and document forensics, with 10 distinct coauthors and a reproducibility mean of 55.
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.
- Profile is name-resolved from public bibliographic data; 'Yuchen Zhou' is a common name and identity may be ambiguous.
- No institution, country, or ORCID is available, so affiliation and geography are unknown.
- Evidence is limited to 2 arXiv papers reviewed in the Evolve catalog; venue history and a full publication record are not included.
- Output, impact, and activity dimensions lack underlying data and are left unscored; the reproducibility mean of 55 rests on only 2 papers.