NLP researcher at the University of Toronto building LLM-agent benchmarks and systems for financial question answering
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
1
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-13) 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—
The output axis is unscored: while 10 recent works are indexed, the record lacks the complete, disambiguated publication list and venue metadata this axis requires, so overall output volume and venue quality cannot yet be reliably assessed.
Impact—
Impact is unscored: with an h-index of 1 and an i10-index of 1, the citation record is too young to measure. This axis would capture accumulated citations, citing works, and influence on subsequent research.
Focus0/100
Focus is recorded as 0/100, yet the underlying field distribution looks highly concentrated rather than scattered: 0.5 for LLM-agent evaluation / financial NLP and 0.5 for NLP / LLM agents in financial QA — effectively one research area under two near-duplicate labels. Taken at face value the score contradicts this distribution, and it most plausibly reflects label fragmentation in the OpenAlex topic mapping rather than genuinely dispersed work.
Collaboration69/100
The collaboration axis scores 69/100, grounded in 31 distinct coauthors across 10 works in the last three years. The recurring partner cluster includes Tung Sum Thomas Kwok, Xinyu Wang, Guang Cheng, Yanzhang Ma, and Hanwei Wu, alongside a longer tail of one-off collaborators; both catalogued papers are middle-authored team efforts, consistent with a contributor role inside a small, arXiv-first NLP group rather than a solo or PI-led pattern.
Activity70/100
Activity scores 70/100 on a steep upward slope: OpenAlex counts-by-year records 2 works in 2025 against 8 in 2026, a fourfold year-over-year increase, with recent3yWorks at 10. Every catalogued output dates from 2026, and the cadence is arXiv-first. At this career stage the volume is modest in absolute terms, but the trajectory suggests a quickly ramping research program built around financial LLM agents.
Reproducibility60/100
Reproducibility scores 60/100, computed from Evolve reviews of 2 papers (reproMean 60). The picture is split: FinInteract rates high on reproducibility readiness, reflecting a released benchmark and diagnostic protocol, while FINSKILLOPS rates low, with author-built evaluations, judge bias, and unverifiable artifacts cited as limitations.
Research summary
Zhenghan Tai is an early-career NLP researcher at the University of Toronto whose work centers on large language model agents for financial question answering. The OpenAlex record linked to ORCID 0009-0003-3896-5131 indexes 10 works in the last three years — 2 in 2025 rising to 8 in 2026 — with 31 distinct coauthors, an h-index of 1, and an i10-index of 1, a profile typical of a researcher just entering the field. Two 2026 arXiv preprints reviewed by Evolve define the agenda: FinInteract, a benchmark showing that frontier LLM agents elicit clarification for ambiguous financial questions at most 28.9% of the time against a 95% ceiling, and FINSKILLOPS, a self-evolving multi-agent system for SEC filing QA with gated skill maintenance. Topic signals — Topic Modeling (9), Information Retrieval and Search Behavior (6), and Domain Adaptation and Few-Shot Learning (3) — indicate breadth beyond finance, though the field distribution places half of indexed output in LLM-agent evaluation / financial NLP and half in financial-QA agents. Reproducibility assessments average 60 across the two papers but diverge sharply: high for FinInteract, low for FINSKILLOPS. All catalogued outputs are preprints with middle-author positions; no peer-reviewed venue record is yet available.
Collaboration pattern
With 31 distinct coauthors across 10 recent works, Zhenghan Tai works in team-based settings rather than solo. The recurring cluster centers on Tung Sum Thomas Kwok and Xinyu Wang, with Guang Cheng, Yanzhang Ma, Hanwei Wu, and others appearing across projects; both catalogued 2026 papers are middle-authored, non-corresponding contributions, indicating a reliable collaborator role within a compact, arXiv-first financial-LLM group rather than an independent lab lead.
Research keywords
Highlights
FinInteract: measuring how rarely LLM agents ask for clarification in financial QA
FINSKILLOPS: gated skill maintenance for SEC filing QA agents
Timeline
- 2025First indexed works appear: OpenAlex counts-by-year records 2 publications in 2025.
- 2026Output accelerates to 8 indexed works; the recent three-year total reaches 10 works with 31 distinct coauthors.
- 2026Co-authors FinInteract (arXiv), benchmarking clarification and intent integration in ambiguous financial QA; reproducibility readiness rated high.
- 2026Co-authors FINSKILLOPS (arXiv), a self-evolving multi-agent system for SEC filing QA; reproducibility readiness rated low.
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.
- Bibliometrics (h-index, counts by year, topics, coauthor counts) come from a single OpenAlex record (identity: openalex) linked to ORCID 0009-0003-3896-5131; coverage of very recent arXiv work may lag.
- Only two papers carry Evolve reviews, both 2026 arXiv preprints with middle-author positions and no corresponding-author role; no peer-reviewed publication record is available.
- Output and impact radar axes are unscored for insufficient data; the recorded focus score of 0 conflicts with a highly concentrated field distribution and may reflect topic-label fragmentation.
- Profile is name-resolved from public bibliographic data; other researchers named Zhenghan Tai could be conflated in the underlying record.