Abhisek Dash

Max Planck Institute for Software Systems

Unclaimed · AI-structured profileORCID OpenAlex

Max Planck Institute for Software Systems researcher auditing content moderation and LLM agents for web search.

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—

Publication-volume and authorship-role metrics are not yet available for this profile, so output cannot be scored. This axis would measure papers per year, venue tiers, and first/corresponding-author share across the full publication record rather than only the reviewed catalog.

Impact—

Citation counts, h-index-style measures, and downstream uptake are absent from the current snapshot, so impact is not scored. This axis would quantify how often and how widely this researcher's papers are cited, reused, and picked up by follow-up work.

Focus0/100

The focus dimension scores 0/100. The underlying field distribution is split exactly evenly at one third each across natural language processing for content moderation and computational social science, computational social science, and NLP/information retrieval for LLM agents. This points to deliberate, problem-driven breadth — content moderation, platform auditing, and agentic web search — rather than narrow single-track specialization.

Collaboration61/100

Collaboration scores 61/100, grounded in 20 distinct coauthors. Recurring partners include Krishna P. Gummadi, Ingmar Weber, and Abhijnan Chakraborty, plus Ayan Majumdar, Pushpdeep Singh, Sayeh Jarollahi, Shounak Paul, and Vabuk Pahari — a network centered on computational social science and trustworthy platform research. On catalog papers he is last author twice and middle author once, none corresponding.

Activity—

Year-by-year publication cadence is not present in the data, so activity cannot be scored. This axis would track publishing continuity, recent output volume, and research engagement across the career timeline; the catalog itself covers only 2026.

Reproducibility43/100

Reproducibility readiness averages 43/100 over the three reviewed papers. Evolve rates the two Bluesky moderation studies as medium readiness — both build on sizeable artifacts (a 4K labeled Bluesky benchmark; an audit whose recall ground truth rests on 27 posts and coauthor labels) — while the agentic web search measurement is rated low, with judge-model heterogeneity and confounds tempering its conclusions.

Research summary

Abhisek Dash is a researcher at the Max Planck Institute for Software Systems (ORCID 0000-0002-5300-8757) whose work sits at the intersection of natural language processing, computational social science, and information retrieval. The Evolve catalog holds three 2026 arXiv papers with him in central authorship roles: last author on a first large-scale audit of Bluesky's content moderation service, last author on an evaluation of whether foundation models can moderate online content, and middle author on the first cross-platform measurement of agentic web search by conversational LLM agents — none as corresponding author. A consistent thread is auditing and benchmarking socio-technical AI systems in production: the Bluesky audit surfaces how a decentralized platform automates moderation and where harms escape it, while the foundation-model study shows vision-language models nearly triple moderation F1 (0.60 vs 0.22) on a 4K Bluesky benchmark. His bibliometric footprint spans 20 distinct coauthors and three equally weighted field clusters, and his collaboration score (61/100) reflects steady work with Krishna P. Gummadi, Ingmar Weber, and Abhijnan Chakraborty, among others. Reproducibility readiness is moderate (43/100 across three papers), with judge-model confounds and co-author-only labels flagged by Evolve reviewers. Output, impact, and activity metrics are not yet available.

Collaboration pattern

Dash's collaboration score is 61/100 with 20 distinct coauthors. His frequent-coauthor list is anchored by Krishna P. Gummadi, Ingmar Weber, and Abhijnan Chakraborty — a computational-social-science and platform-governance cluster — alongside Ayan Majumdar, Pushpdeep Singh, Sayeh Jarollahi, Shounak Paul, Vabuk Pahari, Mahsa Amani, Ines Abdelaziz, and Seungeon Lee. On the three catalog papers he is last author twice and middle author once, none as corresponding author, suggesting contribution-heavy roles within larger MPI-SWS-led teams.

Research keywords

content moderationcomputational social sciencedecentralized social mediaplatform auditvision-language modelsLLM agentsconversational web searchbenchmark constructionin-context learningprompting paradigmshuman-AI collaborationAT Protocol

Highlights

First large-scale audit of Bluesky's content moderation service
The study characterizes how Bluesky automates moderation over the AT Protocol, mapping the landscape of harms that escape automation; Evolve calls the transparency findings novel and convincing, while noting that recall rests on 27 posts and co-author ground truth.
Vision-language models nearly triple Bluesky moderation F1 (0.60 vs 0.22)
On a 4K Bluesky benchmark, the paper compares instruction-driven versus example-driven policy operationalization and finds foundation models reach 0.60 F1 against the platform's 0.22; Evolve notes labels come only from co-authors and no significance tests are reported.
First cross-platform measurement of agentic web search
The study characterizes how conversational LLM agents search the web across four platforms, tracing search decisions, strategies, results, and responses through the agent lifecycle; Evolve flags judge-model heterogeneity and confounds, and rates reproducibility as low.

Timeline

2026Last-author evaluation of foundation models for content moderation shows vision-language models nearly triple Bluesky's moderation F1 (0.60 vs 0.22) on a 4K Bluesky benchmark.
2026Last-author audit of Bluesky's content moderation service appears on arXiv, characterizing the automation of service and the landscape of harms on a decentralized platform.
2026Middle-author cross-platform measurement of conversational LLM agents performing web search is released, covering search decisions, strategies, results, and responses across four platforms.

Papers in the Evolve catalog

Frequent co-authors

Krishna P. Gummadi×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 scores and bibliometric counts derive from a public bibliometric snapshot; citation counts, h-index, and venue-level impact are not available in this profile.
All three catalog papers are 2026 arXiv preprints; peer-review venue status is not reflected in the data.
Output, impact, and activity axes are unscored because the underlying per-year productivity and citation data are missing.
Identity is name-resolved from public bibliographic data (ORCID 0000-0002-5300-8757); the name may be shared by multiple researchers.