Pranab Mohanty

Capital One (United States) · United States

Unclaimed · AI-structured profileOpenAlex

Capital One ML researcher focused on evaluation in explainable recommendation, graph learning, and LLM safety

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

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—

Output is not yet scored for this profile: the OpenAlex snapshot lacks a complete, disambiguated yearly production series, so no 0-100 output score can be computed. This axis would measure publication volume and consistency across the whole career.

Impact—

Impact is not yet scored: the recorded h-index of 7 and i10-index of 4 are too sparse to benchmark against field norms without full citation vectors. This axis would measure citation-based influence, uptake, and the reach of published work.

Focus0/100

The focus score is 0, reflecting a sharply split portfolio: the field distribution is an even 50/50 between explainable-AI recommender systems and graph representation learning, while OpenAlex topics span topic modeling (10), face and expression recognition (9), biometric identification and security (9), face recognition and analysis (9), and explainable AI (5). The 2002-2008 biometrics era and the 2026 evaluation agenda read as two distinct research programs.

Collaboration58/100

Collaboration scores 58. The record lists 17 distinct coauthors with a recurring core of twelve, including Giri Iyengar, Qilong Wu, Sahil Wadhwa, Varun Chandrasekaran, and Dwipam Katariya. Across all three 2026 catalog papers Mohanty holds a middle or last author position and is never the corresponding author, consistent with a senior technical contributor in a stable industry group rather than a solo project lead.

Activity70/100

Activity scores 70. The OpenAlex counts-by-year series shows 10 works in 2026 — the highest single-year total in the record — and 10 works over the last three years, following sparse output in 2019 (1), 2021 (2), and 2023 (1). All three Evolve-catalogued papers are 2026 arXiv preprints, confirming a strongly front-loaded recent burst of activity.

Reproducibility78/100

Reproducibility scores 78, computed as the mean across 2 papers with repro-readiness assessments. Within the Evolve catalog, QUARTET is rated high readiness, while the temporal-attribution benchmark and the jailbreak-evaluation paper are rated medium — the former limited by 100-user sampling and internal inconsistencies, the latter by unadjudicated labels.

Research summary

Pranab Mohanty is a US-based industry researcher at Capital One whose OpenAlex-indexed record falls into two phases. The early phase (2002-2008) concentrates on topic modeling, face and expression recognition, and biometric identification and security — among the record's most frequent topics — and is followed by a 2009-2018 span with no indexed works. Indexed publishing resumed sparsely (1 work in 2019, 2 in 2021, 1 in 2023) and peaked at 10 works in 2026, matching a recent three-year total of 10. Three 2026 arXiv papers in the Evolve catalog define the current agenda, all centered on evaluation methodology: a systematic benchmark of ten explainable-AI methods for temporal attribution in sequential recommendation; QUARTET, a quad-branch cross-attention graph transformer reported to beat RelGT on 8 of 12 RelBench tasks; and SEAV, a validity-aware framework for judging LLM jailbreak evaluations. Across these papers Mohanty is a middle or last author and never the corresponding author, consistent with a senior contributor inside a network of 17 distinct coauthors. Reported citation metrics are modest (h-index 7, i10-index 4), and reproducibility across two assessed papers averages 78.

Collaboration pattern

Mohanty publishes with a stable industry co-author pool: 17 distinct coauthors overall and a recurring twelve-person core — Giri Iyengar, Qilong Wu, Sahil Wadhwa, Varun Chandrasekaran, Akash Pandey, Kanisha Shah, Addrish Roy, Dwipam Katariya, Hongyangyang Shi, Amanda Ding, Kalanand Mishra, and Kyaw Hpone Myint. On all three 2026 catalog papers he takes a middle or last author position and is never the corresponding author, fitting a senior contributor pattern within team-produced benchmark and evaluation work.

Research keywords

Explainable AISequential recommendationTemporal attributionFaithfulness evaluationRelational deep learningGraph transformerRelBenchPersonalized PageRankLLM jailbreak evaluationLLM-as-a-judgeIntegrated gradientsSHAP

Highlights

QUARTET: quad-branch cross-attention graph transformer for relational graphs
QUARTET combines RelGT with causal-PPR sampling and four-branch global attention, beating RelGT on 8 of 12 RelBench tasks; Evolve notes tuning asymmetry and thin margins temper SOTA claims, and rates reproducibility readiness high.
SEAV: validity-aware jailbreak evaluation for LLMs
SEAV adds retrieval-grounded step and order checks to jailbreak scoring; Evolve finds the evidence that prior metrics overcount attack success suggestive, but resting on unadjudicated labels, with medium reproducibility readiness.
Systematic benchmark of XAI methods for temporal attribution in sequential recommendation
This benchmark evaluates 10 explainable-AI methods for temporal attribution in sequential recommender systems; Evolve calls it mostly convincing but weakened by 100-user sampling and internal inconsistencies, with medium reproducibility readiness.

Timeline

2002First indexed work appears in the OpenAlex record (1 work), opening the early phase on topic modeling, face recognition, and biometric security.
2006Early output peaks: 4 indexed works in 2006, within a 2006-2008 cluster totaling 10 works aligned with the record's topic-modeling and biometric topics.
2008Last year of the early cluster (3 works); the following decade, 2009-2018, shows no indexed works in OpenAlex.
2019Indexed publishing resumes with 1 work, beginning the return to research activity after the gap.
2023Sparse ramp-up continues (2 works in 2021, 1 in 2023), preceding the 2026 surge.
2026Record year with 10 indexed works, including all three Evolve-catalogued arXiv preprints: the temporal-attribution XAI benchmark, QUARTET, and SEAV jailbreak evaluation.

Papers in the Evolve catalog

Frequent co-authors

Giri Iyengar×2 Qilong Wu×1 Sahil Wadhwa×1 Varun Chandrasekaran×1 Akash Pandey×1 Kanisha Shah×1 Addrish Roy×1 Dwipam Katariya×1 Hongyangyang Shi×1 Amanda Ding×1 Kalanand Mishra×1 Kyaw Hpone Myint×1

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.

Bibliometric figures (h-index 7, i10-index 4, counts by year, topics, 17 distinct coauthors) come from an OpenAlex snapshot; no ORCID was available, so this profile is name-resolved from public bibliographic data and may merge or split namesakes.
The absence of indexed works in 2009-2018 reflects OpenAlex coverage; unpublished or non-indexed industry work during that period cannot be ruled out.
Output and impact dimensions are unscored due to insufficient data; citation counts for the 2026 arXiv preprints are likely still lagging.
The reproducibility score averages only 2 assessed papers; one of the three catalog papers has no repro-readiness score.