Capital One researcher advancing relational graph transformers and validity-aware LLM jailbreak evaluation
2
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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 output score is available. The indexed OpenAlex record shows h-index 0 and i10-index 0 with no publication-volume aggregates, so the record is too thin to score. This axis would measure cumulative publication volume and productivity over time.
Impact—
Impact cannot be scored yet. Citation-based signals are absent — h-index 0, i10-index 0, and no citation aggregates in the indexed record. This axis would measure citation accumulation, per-year influence, and uptake of the two 2026 preprints as they age.
Focus100/100
The focus axis scores 100 because the OpenAlex field distribution collapses to a single field — machine learning (graph representation learning / relational deep learning) — for the indexed work. Topic counts are broader: information retrieval and search behavior (3) and adversarial robustness (3) lead, with topic modeling, recommender systems, and software engineering research at 2 each. Concentration therefore rests on a thin record rather than a long single-track career.
Collaboration48/100
The collaboration axis scores 48 from a modest footprint: 10 distinct coauthors appear across the indexed record. Repeated partners include Pranab Mohanty, Qilong Wu, Sahil Wadhwa, Varun Chandrasekaran, Kyaw Hpone Myint, Nan Jiang, Xiang Li, Zhe Wu, and Alexandre G. R. Day, suggesting small recurring teams around the 2026 arXiv work rather than a broad cross-institution network.
Activity57/100
Activity scores 57, supported by 6 works in the last three years, all clustered in 2026 per the counts-by-year series (1998: 1; 2026: 6). The cadence reflects a recent, concentrated burst of arXiv preprints rather than sustained multi-year output, so the mid-range score is limited by a short observation window.
Reproducibility100/100
Reproducibility scores 100, but from a base of just one scored paper (reproMean 100 across 1 paper). QUARTET carries high reproducibility readiness in Evolve's review, while the SEAV jailbreak-evaluation paper is rated medium; with only one paper in the repro sample, the perfect score should be read as promising rather than proven.
Research summary
Giri Iyengar is a US-based researcher at Capital One whose recent public output centers on two 2026 arXiv preprints reviewed in the Evolve catalog. The first, QUARTET, introduces a quad-branch cross-attention graph transformer with causal personalized-PageRank sampling and random-walk traces for relational deep learning; Evolve's review finds it beats RelGT on 8 of 12 RelBench tasks while cautioning that tuning asymmetry and thin margins temper SOTA claims. The second, SEAV (Validity-Aware Jailbreak Evaluation), adds retrieval-grounded step- and order-validity checks to LLM jailbreak scoring, with evidence that prior metrics overcount attack success — suggestive, though resting on unadjudicated labels. OpenAlex records 6 works in the last three years (6 in 2026) across a topic profile led by information retrieval and search behavior (3) and adversarial robustness (3), with topic modeling, recommender systems, and software engineering research at 2 each. The indexed h-index and i10-index are both 0, so citation impact cannot be scored. This profile is name-resolved from OpenAlex bibliographic data; no ORCID was supplied, and a lone 1998 count suggests possible record merging. Co-authorship spans 10 distinct collaborators, including Pranab Mohanty, Qilong Wu, and Sahil Wadhwa.
Collaboration pattern
With 10 distinct coauthors in the indexed record, Iyengar's pattern is small recurring teams on 2026 arXiv work. Frequent partners include Pranab Mohanty, Qilong Wu, Sahil Wadhwa, Varun Chandrasekaran, Kyaw Hpone Myint, Nan Jiang, Xiang Li, Zhe Wu, and Alexandre G. R. Day. Both cataloged papers are multi-authored with Iyengar in non-corresponding positions (last on QUARTET, middle on SEAV), consistent with a contributing role in his Capital One research group.
Research keywords
Highlights
QUARTET: wins on 8 of 12 RelBench tasks against RelGT
SEAV: validity-aware jailbreak evaluation for LLMs
Timeline
- 2026QUARTET preprint released on arXiv, introducing a quad-branch cross-attention graph transformer with causal personalized-PageRank sampling and random-walk traces for relational graphs.
- 2026Evolve review of QUARTET: wins on 8/12 RelBench tasks versus RelGT, with high reproducibility readiness but thin margins.
- 2026SEAV preprint released on arXiv, adding retrieval-grounded step- and order-validity checks to LLM jailbreak evaluation.
- 2026Evolve review of SEAV: prior jailbreak metrics may overcount attack success; medium reproducibility readiness.
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 are computed from OpenAlex only; no ORCID was supplied, so this profile is name-resolved from public bibliographic data and may merge or split records.
- The indexed h-index and i10-index are 0 and no citation aggregates exist, so the output and impact axes are left unscored.
- A single 1998 count alongside 2026 (6 works) suggests possible identity merging or sparse coverage; the timeline is restricted to the two 2026 cataloged papers.
- The reproducibility score rests on one paper; the second cataloged paper is rated medium reproducibility readiness.