Motivation
Currently, the grounding workflow requires a specific parent_id or multiple to be selected. However, we could also do lexical/semantic search and then resolving to standard concepts (or just accepting standard concepts during the lookup) without the ancestry.
The grounding would then be a bit more loose and solely rely on the textual similarity as a scoring but could be interesting, especially for MCP/ACP graph exploration.
Pitch
This change requires a re-thinking in the logic of:
- how concepts are resolved and grounded
- the scoring as it currently relies also on the distance to the
parent_id
Alternatives
This change requires a change in the code. There are no alternative except rejecting this feature request alltogether.
Additional context
No response
Motivation
Currently, the grounding workflow requires a specific
parent_idor multiple to be selected. However, we could also do lexical/semantic search and then resolving to standard concepts (or just accepting standard concepts during the lookup) without the ancestry.The grounding would then be a bit more loose and solely rely on the textual similarity as a scoring but could be interesting, especially for MCP/ACP graph exploration.
Pitch
This change requires a re-thinking in the logic of:
parent_idAlternatives
This change requires a change in the code. There are no alternative except rejecting this feature request alltogether.
Additional context
No response