Extending Metadata Filtering concept to all assets

[marcel-vesely-kinit]

This task consists of multiple large sub-tasks each focusing one of the aspects necessary for consolidating metadata filtering feature as a whole:

• Use of an LLM to extract metadata from their JSON format
  • Currently, in the PoC (Proof of Concept) implementation of Metadata filtering, we intentionally omitted this step and only extracted metadata associated with assets whose JSON schema is consistent and contains relevant fields to extract metadata from (HuggingFace datasets)
• Use of an LLM to parse user queries
  • This is mostly done, albeit the implementation is currently tied to dataset asset type only
• Extend Metadata filtering feature to any AIoD assets
  • Each asset type can have its own unique set of fields we can use for metadata filtering purposes