Advanced
Late-Interaction Retrieval (ColBERT)
A bi-encoder crushes a document into one vector and loses which words carried the meaning. A cross-encoder keeps everything but cannot precompute, so it only reranks. Late interaction keeps one vector per token and compares them at query time, buying most of the accuracy of the second at roughly the scan cost of the first, and paying for it in storage.
Get full Premium access · ₹2,000 / $25
Every answer, concept and course, every paper breakdown, all published video lessons, hands-on FDE Lab missions, Premium PDF guides and companion files, the full practice-test bank and work-sample downloads. Referral Premium excludes guide PDFs and their companion files.
6 months · One payment · No auto-renewal
Study alongside free video lessons.
Related concepts
Practise this in questions
RAG & Agent System DesignHow would you chunk documents for retrieval, and how do you know your chunking is good? →easy · FreeRAG & Agent System DesignYou added a cross-encoder reranker. p95 latency roughly doubled and answer quality did not move. Explain what happened and fix it. →hard · PremiumRAG & Agent System DesignWalk me through a RAG pipeline end-to-end, and tell me where it usually breaks →easy · FreeRAG & Agent System DesignKeyword search vs vector search: what does each actually buy you in a RAG system? →easy · FreeLLM & GenAI FundamentalsWhat are embeddings, and how do they make semantic search work? →easy · FreeLLM & GenAI FundamentalsWhy is the dot product the similarity score in attention and embeddings, and when should you normalize to cosine? →medium · Free