Dennis Ng — NLP & AI agent safety: calibrated small-language-model scam screening and agentic PDF-forgery evaluation
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 data is available yet. This axis would measure total paper counts, venue spread, and yearly productivity; only the two cataloged arXiv papers are on record, so no score is computed.
Impact—
Citation-based impact cannot be scored: the input carries no citation counts, h-index, or venue-level influence signals. This axis would measure citations, downstream uptake, and scholarly influence; it remains unscored until such data is linked.
Focus0/100
The focus score is 0, derived from an even 50/50 field distribution: 'AI agent safety and document forensics' at 0.5 and 'natural language processing (NLP)' at 0.5. Rather than signaling scattered work, this shows two equally weighted research lines — calibrated scam-call screening with small language models, and measurement of agentic PDF forgery — with no dominant specialization visible yet in the record.
Collaboration48/100
The collaboration score is 48, computed from a record showing 10 distinct coauthors. The frequent-collaborator list is led by Simiao Ren and Ankit Raj — each listed twice, indicating repeat co-authorship — alongside Tommy Duong, Yuxin Zhang, Xingyu Shen, Kidus Zewde, Yuchen Zhou, and Neo Tiangratanakul. Both cataloged papers carry middle authorship positions with no corresponding role, consistent with a contributor inside a compact team rather than a solo or leading-authorship pattern.
Activity—
No temporal signals exist beyond the two 2026 paper years — no submission cadence, career timeline, or yearly volume. This axis would measure publishing rhythm and recent-output trends; the data is insufficient to score it.
Reproducibility55/100
Reproducibility scores 55 — a mean (reproMean 55) across 2 scored papers, both rated 'medium' readiness. The scam-screening work is candid about test-set exposure but rests entirely on synthetic data; the PDF-forgery study applies a rigorous rule-based scoring design whose rankings are confounded by server-side heterogeneity. Together they indicate documented, partially reproducible artifacts rather than fully turnkey replication packages.
Research summary
Dennis Ng is an early-stage researcher whose visible output sits at the intersection of natural language processing and AI agent safety, with the cataloged field distribution split evenly (0.5 / 0.5) between the two. The Evolve catalog contains two 2026 arXiv papers, in both of which Ng holds a middle authorship position rather than first or corresponding author. The first, on CallScreenBench, shows a JevLite typed-decision readout fine-tuned into Qwen3-4B delivering scam-screening judgments non-inferior to an LLM judge at roughly 30x lower latency, with candid treatment of test-set exposure — though all evaluation data is synthetic. The second measures how reliably LLM agents forge archived PDFs against a non-agentic control, reporting an 81.1% raw forgery success rate that falls to 46.2% under strict filtering, using a rigorous rule-based scoring design whose rankings are nonetheless confounded by server-side heterogeneity. Public bibliometric records indicate about 10 distinct coauthors and a reproducibility readiness mean of 55 across the 2 scored papers. Beyond these records, no institution, country, or career information is available in the input data.
Collaboration pattern
The record shows 10 distinct coauthors and a collaboration score of 48. The frequent-coauthor list centers on Simiao Ren and Ankit Raj — each listed twice, implying sustained repeat collaboration — together with Tommy Duong, Yuxin Zhang, Xingyu Shen, Kidus Zewde, Yuchen Zhou, and Neo Tiangratanakul. With middle authorship on both cataloged papers and no corresponding-author role, the pattern reads as a consistent team contributor within a stable, moderately sized group rather than an independent lead.
Research keywords
Highlights
One-pass scam screening at ~30x lower latency
Quantifying how reliably agents forge archived PDFs
A measurement-and-calibration methodology across both lines
Across the two cataloged papers, the common thread is careful evaluation design — typed-decision readouts with calibration on one side, rule-based verification on the other — both rated medium reproducibility readiness in the Evolve catalog.
Timeline
- 2026Enters the Evolve catalog with the CallScreenBench study: JevLite typed-decision readout on Qwen3-4B screens scam calls in one pass at ~30x lower latency than an LLM judge.
- 2026Publishes the agentic document-forensics measurement: LLM agents forge archived PDFs at 81.1% raw success (46.2% under strict filtering) against a non-agentic control.
- 2026Bibliometric record links about 10 distinct coauthors across the two 2026 papers, with Simiao Ren and Ankit Raj appearing as repeat collaborators.
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.
- "Dennis Ng" is a common name; this profile is name-resolved from public bibliographic data only, with no ORCID or institutional linkage in the input.
- Bibliometric scores are derived from OpenAlex-based public records; output, impact, and activity axes are unscored because the underlying metrics are unavailable.
- Coverage is limited to two 2026 arXiv preprints with medium reproducibility readiness; no peer-reviewed venue record is included.
- Both papers list Ng in a middle authorship position, so first/corresponding-author conclusions should not be drawn from this data.