CustomIR: Unsupervised Fine-Tuning of Dense Embeddings for Known Document Corpora

Nathan Paull

Dense embedding models have become critical for modern information retrieval, particularly in RAG pipelines, but their performance often degrades when applied to specialized corpor…