Fine-tuning

Fine-tuning means continuing the training of an existing model on a targeted corpus — internal case law, customer support history, technical documentation — to improve its performance on that specific domain.

For most business uses, RAG is enough and easier to maintain (knowledge updates by re-indexing documents, no retraining). Fine-tuning remains valuable for specializing a local open-source model on domain vocabulary or a highly constrained output format.