What is the Spring AI ETL pipeline and how does it work?

The Spring AI ETL (Extract-Transform-Load) pipeline is a composable data processing abstraction for building RAG ingestion workflows. Rather than wiring readers, splitters, and vector stores manually in imperative code, ETL lets you declare a pipeline as a chain of typed transformations that process List<Document> at each stage.

The three pipeline roles map directly to ETL concepts:

• DocumentReader — Extract: reads source documents and returns List<Document>.
• DocumentTransformer — Transform: a function that takes List<Document> and returns a (modified) List<Document>. TokenTextSplitter, MetadataEnricher, and ContentFormatTransformer all implement this interface.
• DocumentWriter — Load: consumes List<Document> and persists them. VectorStore implements DocumentWriter.

// Functional pipeline style
DocumentReader reader = new PdfDocumentReader(resource);
DocumentTransformer splitter = new TokenTextSplitter();
DocumentTransformer enricher = new KeywordMetadataEnricher(chatModel, 5);
DocumentWriter store = vectorStore;

// Chain and run
store.accept(
    enricher.apply(
        splitter.apply(reader.get())));

Because DocumentTransformer is a standard Java Function<List<Document>, List<Document>>, you can compose transformers using Function.andThen(). This makes it straightforward to add steps like metadata enrichment, deduplication, or content filtering anywhere in the chain without restructuring the rest of the pipeline.