Is Lorry just one big language model?

No. She is a system of 30+ specialized models, most of them trained on your data. The small, fast ones handle detection, ranking, and risk scoring in house: reading intent, scoring RFE risk before you file, flagging hostile messages, hearing stress in a caller's voice. The big language model only runs when a task actually needs reasoning, which keeps her fast and cheap. She sharpens every week as the models retrain on your corrections.