Positioning This entry points to a subdirectory of Milvus's official bootcamp repository, bootcamp/RAG/search_with_jev: nine notebooks on retrieval-augmented search in which Jev supplies the judgment calls. The bootcamp is the project's tutorial repository, and pymilvus is its companion Python package. Together the notebooks form a practical cookbook for the pattern, one recipe per decision point.
What it does The notebooks cover reranking, context filtering, deciding when to stop searching, graph-relationship reranking, query routing, cache reuse, pre-indexing selection, injection screening, and evidence review — the decision points inside a RAG pipeline where a calibrated judgment is an alternative to generation. The collection spans the retrieval flow end to end, from pre-indexing selection through query routing to evidence review. In each, Jev's role is the judgment call rather than generating text. Together the notebooks form a complete, runnable pipeline combining Gemini embeddings with Jev judgments, usable both as tutorials and as copy-paste starting points. The runnable pipeline combines Gemini embeddings with Jev judgments end to end.
Characteristics The JevRerankFunction integration has been merged into pymilvus[model], so Jev-based reranking is available directly to pymilvus users without extra integration work. As official vendor tutorial material, each of the nine decision points comes with runnable code.
When to use For developers adding decision points to a RAG pipeline: read the notebooks as tutorials or lift the code as a starting point. It is also vendor documentation of Jev being used as a component in a retrieval stack. The material is organized around the Milvus and pymilvus stack, so porting it to another retrieval stack means adapting the code yourself.