Resources

GitHub Repos

ECHO: Encoding Communities via High-order Operators: (Feb 2026 release) ECHO, a scalable, self-supervised framework for attributed community detection in large-scale networks.

Generalized Louvain Method: RECOMMENDED 2026 new scalable implementation in Python (works for large scale networks up to millions of nodes)

GEM: A Python library for Graph Embedding Methods

From Bots to Agents: A curated index of the research behind the lecture: bot detection and its limits, coordinated inauthentic behavior, influence operations, and the shift from scripted automation to LLM-driven agents. 

AI Research Collection: A curated index of my research on artificial intelligence: model safety and deceptive behavior, auditing and bias, agentic systems, generative-AI risk, and evaluation.

Benchmarks

OWMI: (2026) Open-Weight Masked Introspection, a framework for measuring what a language model can report about its own computation. It intervenes on residual-stream sites, attention heads, and sparse-autoencoder features, then scores the model’s report against sham runs, impact-matched random perturbations, and a text-only observer. Harness, batteries, and model card on Hugging Face; paper.

QuantiBias: (2026) A benchmark for quantization-induced bias in LLMs. It pairs a multilingual generative stereotype probe with the refusal and multiple-choice controls that isolate open-ended generation, and rates the severity of what the model produces. Harness, judge rubrics, and reconstruction manifests on Hugging Face; paper.

MOSAIC: (2026) A benchmark for the moral, social, and individual value dimensions of LLMs, built from nine validated psychology instruments and four ethical-dilemma scenario sets. Dataset on Hugging Face, code on GitHub; paper.