ALucek/rag-reranking

Implements cross-encoder and late interaction reranking architectures that directly compare queries against retrieved passages for fine-grained relevance scoring, addressing the limitation that initial embedding-based retrieval may return semantically similar but contextually suboptimal results. Provides practical comparisons of popular reranking approaches to optimize the second stage of RAG pipelines, with focus on improving retrieval quality beyond vector similarity matching.

No commits in the last 6 months.