I build language models, datasets and benchmarks for low-resource languages and historical text, with a focus on getting strong results from small models and little data. I care about evaluation that holds up when there is no ground truth, and I release everything openly with full provenance.