Visualizers
Learn by moving things
Interactive explainers for the ideas behind retrieval, embeddings, and agents. A formula tells you what is true; dragging a vector shows you why. Each one also appears inside the article it belongs to, and runs entirely in your browser, so drag, poke, and break it.
Deleting the file, not just the original
From: RAGOps: RAG Is a Distributed System, Not a Demo →When a source document is deleted, its chunks are still in the index. A broken pipeline leaves them there, so the model keeps citing deleted data; a correct one purges every chunk whose parent is that document, in the same operation.
A customer record is one source document, chunked into 5 pieces months ago and stored in the index.
Measuring with the same ruler
From: RAGOps: RAG Is a Distributed System, Not a Demo →Query and document must be embedded with the same model to be comparable. Swap the embedding model and old vectors live in a different space from new queries, so similarity scores become meaningless: you have to re-embed the whole corpus.
Query and document are both embedded with model v1, the same ruler. They land close together, so the match is real.
A recorder for every answer
From: RAGOps: RAG Is a Distributed System, Not a Demo →A wrong answer is not debuggable as a whole. A per-request trace follows the query through query-processing, retrieval, rerank, and generation, recording the numbers at each hop, so you can localize the failure to one step instead of guessing.
A user reports a wrong answer. 'The bot was wrong' is not debuggable. Open its trace instead.