1. Applications Administration Guide
  2. Troubleshooting and Best Practices

Troubleshooting and Best Practices

Symptom Checks
RAG UI or buttons are not visible Validate RAG configuration, manifest expressions, renderer file mappings, view ownership and responsibilities, and browser cache behavior.
Similar SR results are missing or incomplete Confirm that OpenSearch is enabled, the Service Requests – Modern Search category is indexed, and all required SR fields are present in the indexed content.
Cloud RAG response is not returned Validate OCI connectivity and configuration, including authentication settings, compartment ID, region, key file, bucket/knowledge base setup, agent endpoint, and network access.
Attachment context is not included Verify Service Request Attachment settings, file accessibility, SummarizeAttachmentChat user properties, and support for the required document and image formats.
Responses are generic or not grounded Review prompt templates, chunking strategy, embedding model alignment, result limits, hybrid search weighting, and data freshness to ensure optimal retrieval quality.

Best Practices

  • Security and Governance
    • Index only data that users are authorized to access. Since RAG does not enforce Siebel data access controls, customer security requirements, user responsibilities, and data retention policies must be taken into account when selecting data sources.
  • Knowledge Base Configuration
    • Use the same embedding model for both design-time ingestion and runtime query embedding within a given knowledge base to ensure embedding consistency.
    • Store structured data as metadata or filters, while embedding descriptive text fields to support effective semantic search.
    • Evaluate and optimize chunk size, overlap, and contextual chunking configurations for large documents to enhance retrieval accuracy and relevance.
  • Search Optimization
    • Fine-tune hybrid search parameters, including search weights, result limits, and prompt augmentation templates, in a non-production environment before deployment.

  • Monitoring and Maintenance

    • Continuously monitor RAG-generated responses for accuracy, completeness, and proper grounding against source content. Refine data preparation processes, prompts, and retrieval configurations as needed.
    • Re-index or refresh the knowledge base whenever significant source data changes occur or when additional data categories are introduced.