Krishna P. Gummadi

Max Planck Institute for Software Systems

Unclaimed · AI-structured profileORCID OpenAlex

Empirical auditor of platform governance and LLM-mediated web search at Max Planck Institute for Software Systems

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-16) 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—

Insufficient data: the snapshot holds no publication counts, so output is unscored. This axis would measure the volume and cadence of papers and preprints over time, enabling comparison against field norms.

Impact—

No citation or influence indicators are available in this snapshot, so impact is unscored. This axis would capture citations, downstream uptake, and scholarly reference to the researcher's work once bibliometric coverage is complete.

Focus0/100

The focus score of 0 reflects an unusually even spread in the raw field distribution: natural language processing for content moderation, computational social science, and NLP/information retrieval for LLM agents each account for exactly one third of the cataloged work. The portfolio reads as deliberately broad across three adjacent strands that nonetheless converge on platforms, moderation, and language-model-mediated systems.

Collaboration61/100

The collaboration score of 61 rests on 20 distinct coauthors recorded in the snapshot. The frequent-collaborator list — Abhisek Dash, Ayan Majumdar, Pushpdeep Singh, Sayeh Jarollahi, Ingmar Weber, Shounak Paul, Ines Abdelaziz, Seungeon Lee, Vabuk Pahari, Abhijnan Chakraborty, Mahsa Amani — points to stable, repeat empirical teams. All three cataloged 2026 papers place this researcher in a middle position without corresponding role, consistent with contribution inside team-led audits.

Activity—

Recent-publication cadence cannot be assessed from the snapshot, so activity is unscored. This axis would track the rate and recency of new outputs across successive years, indicating whether the research program is currently accelerating or slowing.

Reproducibility43/100

The reproducibility score of 43 is the mean (reproMean 43) over three reviewed papers (reproPapers 3). Evolve rates the Bluesky moderation audit and the foundation-model moderation benchmark at medium readiness, while the cross-platform agentic web search study is rated low, owing to judge-model heterogeneity and confounds. Ground-truth labels built only by coauthors further constrain independent replication.

Research summary

Krishna P. Gummadi is a researcher at the Max Planck Institute for Software Systems whose Evolve catalog centers on the empirical study of online platforms and the language models increasingly embedded in them. Three 2026 arXiv papers, all listing this researcher in a middle authorship position, define the current profile. One line of work audits decentralized social media: a first large-scale characterization of Bluesky's content moderation service built on the AT Protocol, and a companion evaluation asking whether foundation models can moderate content, where vision-language models nearly triple Bluesky's moderation F1 (0.60 versus 0.22) on a 4K-post Bluesky benchmark under instruction- versus example-driven policy operationalization. A second strand measures agentic web search, providing the first cross-platform characterization of conversational LLM agents across four platforms, from search decisions and strategies through results and responses. The radar records 20 distinct coauthors, a collaboration score of 61, and a reproducibility score of 43 averaged over three papers, with the field split evenly across NLP for content moderation, computational social science, and LLM-agent information retrieval. Output, impact, and activity axes lack sufficient data and are left unscored. Evolve's reviews credit the transparency findings while flagging limits: recall resting on 27 posts, coauthor-only ground-truth labels, missing significance tests, and judge-model heterogeneity.

Collaboration pattern

The snapshot records 20 distinct coauthors and a collaboration score of 61. Frequent collaborators include Abhisek Dash, Ayan Majumdar, Pushpdeep Singh, Sayeh Jarollahi, Ingmar Weber, Shounak Paul, Ines Abdelaziz, Seungeon Lee, Vabuk Pahari, Abhijnan Chakraborty, and Mahsa Amani. All three cataloged 2026 papers place this researcher in a middle authorship position without corresponding role, suggesting contribution within team-led empirical audits rather than solo-led projects.

Research keywords

content moderationdecentralized social mediaBlueskyAT Protocolplatform auditvision-language modelsbenchmarkin-context learningprompting paradigmsLLM agentsweb searchhuman-AI collaboration

Highlights

First large-scale audit of Bluesky's content moderation service
Evolve's review calls this a convincing first large-scale audit of the moderation service behind the decentralized AT Protocol platform Bluesky, surfacing novel transparency findings; recall, however, rests on 27 posts and coauthor-provided ground truth, and reproducibility readiness is rated medium.
Vision-language models nearly triple Bluesky's moderation F1 (0.60 vs 0.22)
On a 4K-post Bluesky benchmark, this study evaluates instruction-driven versus example-driven operationalization of moderation policy with foundation models, reporting F1 of 0.60 against Bluesky's 0.22. Evolve notes the strong study design but co-author-only labels and no significance tests; readiness is medium.
First cross-platform measurement of agentic web search
This characterization studies conversational LLM agents searching the web across four platforms, tracing search decisions and strategies through to results and responses. Evolve highlights the rich lifecycle findings while cautioning that judge-model heterogeneity and confounds temper conclusions; reproducibility readiness is low.

Timeline

2026Evaluates whether foundation models can moderate online content, showing vision-language models reach F1 0.60 versus 0.22 for Bluesky on a 4K-post benchmark.
2026Publishes the first large-scale characterization of Bluesky's content moderation service, from automation of service to the landscape of harms.
2026Releases the first cross-platform characterization of web search by conversational LLM agents, spanning four platforms.

Papers in the Evolve catalog

Frequent co-authors

Abhisek Dash×3 Ayan Majumdar×2 Pushpdeep Singh×2 Sayeh Jarollahi×2 Ingmar Weber×2 Shounak Paul×1 Ines Abdelaziz×1 Seungeon Lee×1 Vabuk Pahari×1 Abhijnan Chakraborty×1 Mahsa Amani×1 Seungeon Lee×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.

Radar dimensions are computed from a limited bibliometric snapshot; the output, impact, and activity axes lack sufficient data and are left unscored.
Only the three 2026 arXiv papers reviewed by Evolve are covered; earlier work and peer-reviewed venues fall outside this snapshot.
All cataloged papers list this researcher as a middle author without corresponding role, so leadership attribution from these records is limited.
The profile is name-resolved from public bibliographic data tied to the listed ORCID and institution; demographic and career details are outside the data scope.