hi, i'm lyra

i like poking at things to see how they work. i build tools to peer under the hood of llm systems: interp, evals, redteaming, base models. i also focus on data, both generation and analysis. visualizations, synthetic data, statistical models, attempting realism. outside of the more serious projects, i like training small models to behave in stupid ways, producing music, categorizing random things, and overengineering shitposts. here are some projects that graduated into a releaseable state: