office-dynamixs-ai/doc/RAG_SERVICE.md

The RAG ClusterService EJB

With the EJB ClusterService clients can index and retrieve embeddings.

Index

A text chunk can be index by:

  clusterService.insertVector(workitem.getUniqueID(),
                            workitem.getModelVersion(),
                            workitem.getTaskID(),
                            chunk,
                            indexResult);

Retrieval

The retrieval can be based on text embedding only or refined by workflow metadata like the modelVersion or the TaskId

   List<RetrievalResult> retrievalResultList = clusterService.searchEmbeddings(embeddings, 5);

You can refine the retrieval by providing additional workflow context

1. Task ID Range: Suche alle Tasks zwischen 1000-1999

List<RetrievalResult> rangeResults = clusterService.searchEmbeddingsWithTaskRange(
    queryEmbedding,
    5,
    null,  // all Model Versions
    1000,
    1999
);

2. Wildcard Model Version: all products by wildcard search

List<RetrievalResult> productResults = clusterService.searchEmbeddingsWithModelWildcard(
    queryEmbedding,
    5,
    "product-de*",  // product-de-1.0.0, product-de-2.0.0, etc.
    null
);

3. Combination products and Task Range 1000-1999

List<RetrievalResult> combinedResults = clusterService.searchEmbeddings(
    queryEmbedding,
    5,
    Arrays.asList("product-de*"),
    null,  // no specific  Task IDs
    1000,  // Task Range Start
    1999   // Task Range End
);

4. Multiple Wildcards: all german models

List<RetrievalResult> germanResults = clusterService.searchEmbeddings(
    queryEmbedding,
    5,
    Arrays.asList("product-de*", "offer-de*", "invoice-de*"),
    null, null, null
);