Document Enrichment Processors

[medcl]

🧠 Document Enrichment Processors

Implement a set of processors that can transform, enrich, or normalize documents after they are fetched by a connector and before indexing.

These processors operate independently of the data source and can be chained in a pipeline to apply transformations such as field renaming, extraction, normalization, tagging, and more.

✅ Objectives

• Define a standard processor interface
• Support common field-level transformations
• Allow flexible and composable enrichment pipelines

🛠️ Example Processor Types

Processor Type	Description
rename_field	Rename fields in the document (e.g., title → doc.title)
extract_regex	Extract substrings using regex from text fields
set_value	Set or override a field with a constant value
timestamp_parser	Convert string timestamps to a unified format
truncate	Limit string or array field lengths
add_tags	Append static or dynamic tags to a document
lowercase	Normalize text to lowercase

🔧 Configuration Example

pipeline:
  - name: enrich_documents
    auto_start: false
    keep_running: true
    processor:
      - consumer:
          auto_commit_offset: true
          queue_selector:
            keys:
              - indexing_documents
          consumer:
            group: enriched_documents
            fetch_max_messages: 10
          processor:
            - document_summarization:
                model: $[[env.ENRICHMENT_MODEL]]
                input_queue: "indexing_documents"
                min_input_document_length: 500
                output_queue:
                  name: "enriched_documents"
                  label:
                    tag: "enriched"

📁 Reference

• https://github.com/infinilabs/gateway/tree/main/pipeline

Some existing legacy code can be migrated to processors: https://github.com/infinilabs/crawler/tree/master/pipeline/joints