Best practices for indexing and ranking knowledge base content for an AI assistant

I’m working on an AI assistant that needs a structured indexing framework for its knowledge base. The initial set includes around 20 resources, but some are quite old and may contain outdated information. These older documents should remain available for contextual purposes but be deprioritised compared to newer, validated content.

The goal is to:

Surface accurate, up-to-date content first while keeping legacy materials accessible when needed.

Use metadata and weighting signals (e.g., freshness, reliability, document type, versioning, deprecation flags) to optimise relevance.

Ensure scalability and maintainability as new resources are added and outdated ones phased out.

Support retrieval-augmented generation (RAG) for accurate, source-cited answers.

What are the recommended best practices, architectures, or tools for designing an indexing and ranking strategy that meets these requirements in an enterprise environment? Any guidance on metadata schema, ranking signals, and approaches for long-term maintainability would be greatly appreciated.