Invited talk at the Massachusetts Institute of Technology Lincoln Lab; discussed information and user modeling in fragmented digital ecosystems and how to generalize our understanding of the overarching information ecosystem.

Invited talk at the University of Oxford's OII's Future of Social Media Research Workshop; discussed information and user modeling in fragmented digital ecosystems and how to generalize our understanding of the overarching information ecosystem.

Presented at Meta CSS, in this presentation I showcase a novel framework for modeling cross-platform discourse networks for users across social media platforms. Applied to Truth Social and X, our method finds significant structures that traditional methods miss entirely.

Presented at AAAI ICWSM 2025, we show how war bloggers on Telegram and Gab use moralized language to justify harm, introducing the first scalable framework grounded in sociological theory to detect and analyze "othering" as it unfolds in real time across conflict and non-conflict settings.

Presented at AAAI ICWSM 2025, this introduces a dynamic clustering framework for tracking the real-time evolution of war-related narratives on Telegram, and shows how Russian and Ukrainian communities construct divergent realities through adaptive, internally complex storylines.

Presented for Cohere, I develop a risk-science framework for evaluating and hardening multilingual LLMs. It moves beyond static refusal benchmarks to measure real behavioral risks, including the portability of jailbreaks across languages, the persistence (or decay) of safety patches.

Presented at the AAAI ICWSM 2025 DETOX Workshop, this talk introduces a framework for identifying cross-platform narrative enclaves–semantically coherent, high-echo discourse zones–as structural launchpads for toxic content for the detection of harmful narratives before they go viral.

Presented at Stanford, in this presentation I showcase a novel framework for modeling cross-platform discourse networks for users across social media platforms. Applied to Truth Social and X, our method finds significant structures that traditional methods miss entirely.

This presentation explores how large language models can detect and analyze nuanced forms of othering across social media, revealing how moral and threat-based language is used to justify exclusion, escalate conflict, and shape group identity.

This presentation, delivered at Stanford and USC, offers an in-depth exploration of Otherism—examining its nature, mechanisms, and objectives, as well as outlining the essential work required to address it.