Early-stage researcher in AI governance for agentic LLM systems; first author on compositional policy violations
1
papers here
1
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-21) 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 cataloged record contains only one 2026 first-author preprint, which is too little to compute a stable output measure. This axis would track total publications and venue distribution over time.
Impact—
Not yet scored: no citation counts or downstream-uptake signals are available for the single cataloged paper. This axis would measure citation accumulation, h-index-type indicators, and influence on subsequent work.
Focus100/100
The field distribution is perfectly concentrated: the raw fieldDist attribute assigns 100% of the cataloged output to the single field 'AI governance and safety (agentic LLM systems)', which yields the maximum focus score of 100. With only one cataloged paper, this concentration describes a deliberately narrow entry point into agentic-AI policy compliance rather than a long-term specialization, and it should be re-read as more work accumulates.
Collaboration28/100
The collaboration score of 28 rests on a small raw base: three distinct co-authors appear in the record — Sri Sai Charith Grandhi, Rajesh Gupta, and Sumit Mamoria — the named frequent collaborators on the 2026 first-author arXiv preprint. Working within a compact, stable team is typical of an early independent line of work, but the limited co-author network means external collaboration breadth is not yet established.
Activity—
Not yet scored: only one paper year (2026) is on record, so publication cadence and year-over-year activity cannot be computed. This axis would track annual publishing rate and research continuity across the career.
Reproducibility20/100
Reproducibility is scored at 20, computed as a mean of 20 over exactly one paper with a reproducibility assessment — the 2026 taxonomy paper, whose readiness is classed as low. Evolve's review describes the proposed detection architecture as 'entirely unevaluated', so no benchmarks, baselines, or released artifacts yet support independent replication; empirical validation would most directly raise this score.
Research summary
Ashwini Kurady is an early-career researcher working at the intersection of AI governance and agentic LLM systems, profiled here from public bibliographic data. The Evolve catalog contains one paper: the 2026 arXiv preprint 'Compositional Policy Violations: When Step-Level Compliance Fails In Agentic AI Workflows', on which Kurady is first author. It proposes a four-class taxonomy of compositional policy violations in agentic workflows and pairs it with a detection architecture for runtime policy compliance; Evolve's review calls it a sharp concept that is entirely unevaluated, with low reproducibility readiness. The bibliometric radar reflects this early stage: focus is a perfect 100 because the entire cataloged field distribution sits in AI governance and safety (agentic LLM systems); collaboration scores 28 on a base of three distinct co-authors — Sri Sai Charith Grandhi, Rajesh Gupta, and Sumit Mamoria; and reproducibility averages 20 across one assessed paper. No institution, country, ORCID, citation, or publication-volume data are available in the current snapshot, so output, impact, and activity are unscored. The profile should therefore be read as a directional signal of research interest — policy compliance and runtime verification for multi-agent LLM workflows — rather than an established track record, and it will firm up as further papers are cataloged.
Collaboration pattern
The on-record collaboration pattern is a compact team: three distinct co-authors — Sri Sai Charith Grandhi, Rajesh Gupta, and Sumit Mamoria — appear as frequent collaborators on the 2026 first-author preprint, giving a collaboration score of 28. No cross-institutional or senior-advisor patterns can be inferred from the current snapshot, which lacks affiliation data; the network should widen if subsequent agentic-AI governance work is cataloged.
Research keywords
Highlights
Four-class taxonomy of compositional policy violations
The 2026 first-author arXiv preprint introduces a taxonomy for cases where individually compliant steps compose into policy-violating agentic LLM workflows, giving AI governance a vocabulary for a failure mode that step-level checks miss.
Detection architecture proposed but entirely unevaluated
The paper pairs the taxonomy with a detection architecture for runtime compliance, yet Evolve's review finds it entirely unevaluated and rates reproducibility readiness low, with no benchmarks or released artifacts yet on record.
A sharply focused entry into agentic AI governance
With the entire cataloged field distribution (1 of 1 papers) in AI governance and safety (agentic LLM systems), the focus score reaches 100, and the work is co-produced with a compact three-person collaborator team.
Timeline
- 2026First-author arXiv preprint 'Compositional Policy Violations: When Step-Level Compliance Fails In Agentic AI Workflows' enters the Evolve catalog, the sole paper on record.
- 2026The preprint proposes a four-class taxonomy of compositional policy violations plus a detection architecture; Evolve's verdict: a sharp concept, entirely unevaluated, with low reproducibility readiness.
- 2026Collaborates with Sri Sai Charith Grandhi, Rajesh Gupta, and Sumit Mamoria on the preprint — the three distinct co-authors in the record, forming the current collaboration base.
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.
- Bibliometric scores are computed from public bibliometric data only (OpenAlex-based); no institution, country, or ORCID was available, and the profile is name-resolved from public bibliographic records.
- Output, impact, and activity axes are unscored because the snapshot contains a single 2026 paper with no citation or volume data.
- The focus score of 100 rests on one cataloged paper, so it signals early concentration rather than established specialization.
- The reproducibility score (mean 20) is derived from exactly one assessed paper, rated low for reproducibility readiness.