Can a language-model chatbot clarify the rules of an economics experiment without biasing subjects' choices? Adversarial synthetic subjects attack the bot across thousands of conversations while an independent judge scores every reply. The interactive site explores the agent pipelines, the adversarial subjects, the classifier, real transcripts, the literature, and the results — on canonical strict-judge rates, in the setting of Oprea's (2024) work on choice under risk.
Open project →A community-driven platform for tracking and comparing data quality across online survey platforms. Researchers contribute study results that populate a live, filterable dashboard with metrics on attention, AI and bot detection, and account fraud, with lab and AI benchmarks as reference points. Intended to give survey researchers a standardised, evidence-based view of quality across platforms.
Coming soonA set of interactive mini-games (Steady Hand, Catcher, Pendulum, Chase the Light, Ink, Fire, Plant) that hide survey questions inside playful tasks. An ongoing line of work on how gamified engagement shapes response quality in online surveys, planned as a separate paper.
Open project →A browser-based tool for batch text classification with large language models. Users upload a dataset, define a JSON schema for classes, pick a model, and run classifications with cost estimation, live progress monitoring, and JSONL export. Intended for researchers running text-as-data work at scale.
Open project →