Using Knowledge Transformer
Transform your source documents into AI-optimized knowledge documents, refine them iteratively, and validate their quality.
Use Knowledge Transformer to create and maintain structured taxonomies from your documentation. The transformation process organizes information hierarchically with metadata, enabling AI agents to efficiently search, filter, and retrieve relevant content.
To understand more about the output generated from Knowledge Transformer, see Understanding the output.
Potential knowledge sources
- Subject matter expert (SME) documents, presentations, and interview transcripts
- Technical documentation, manuals, and guides
- Presentation content and training materials
- Video transcripts and meeting recordings
Consider providing multiple complementary documents across successive transformations to enable a broader coverage, as overlapping content can reinforce key concepts.
For more information on the file types supported by Knowledge Transformer, see Supported file formats. Knowledge Transformer will validate file types before transformation.
Generating a taxonomy
Generate a new taxonomy from your source documents using the transform command. The first generation creates the taxonomy structure from scratch, processing your input file and organizing the extracted information into a hierarchical directory structure. You can transform individual files as input for Knowledge Transformer.
For more information, see Generating a taxonomy.
Refining an existing taxonomy
Refine an existing taxonomy to improve and expand it without starting over. Refinement allows you to add new information, fix issues, and enhance coverage while preserving the existing structure. Use the same process as generating a taxonomy, providing your existing taxonomy directory and an additional source document as input. You can refine iteratively multiple times, continuously enhancing your AI-optimized knowledge base.
For more information, see Refining an existing taxonomy.
Validating the quality of a taxonomy
Manually validate the quality of your taxonomy to ensure it meets your requirements. Quality validation helps you assess the completeness, accuracy, and organization of the extracted information before deploying the taxonomy for AI agent use.
Check for terminology accuracy and consistency in your generated taxonomy, especially for content derived from audio transcripts where terms might be misheard.
Compare against source documents and verify that key concepts are captured at a reasonable level of detail for retrieval and reasoning tasks. Understanding any missing key concepts will enable you to identify what sources to provide when iteratively refining your taxonomy.
Ingesting a taxonomy
Ingest your generated taxonomy to make the knowledge available to AI agents. Upload the taxonomy to S3 and use zassist to complete the ingestion process.
For more information, see Ingesting content through CLI.