Beyond the basic pipeline
Better retrieval. Clearer decisions.
Build an advanced RAG pipeline one decision at a time. Short explanations, moving data, and real retrieval results—all in the same Luma Shop example.
Prepare text
Clean source text, find useful boundaries, and recover the context a small passage needs.
3 short lessons02Dense & sparse
Inspect dense coordinates, sparse term signals, and a separate graph-based shortcut for finding vectors.
2 short lessons03Hybrid search
Combine dense and lexical result lists while keeping their different scoring scales separate.
1 focused lesson04Reranking
Inspect how a cross-encoder changes the order of a limited candidate set.
1 focused lesson05Query strategies
Compare rewriting, decomposition, and hypothetical search text without mistaking them for evidence.
1 focused lesson06GraphRAG
Connect products, suppliers, and policies, then inspect the documents supporting each link.
1 focused lesson07Context & evaluation
Pack a concise, traceable context and check retrieval separately from answer support.
2 short lessons08Pipeline lab
Combine the techniques and inspect exactly where the evidence changes.
1 focused lessonRuns offline with recorded MiniLM embeddings and cross-encoder scores. Search, fusion, graph traversal, and context packing run here. Answers are labeled teaching simulations.