Hello!
The answer to this question is two-fold:
a) Converting a source mesh in ARR (fbx or gltf/glb) persists the names of the nodes in the scene graph. If you perform a raycast, you get the hit entity, so you can use its name to identify meta data in your own, client-side data base.
b) ARR also supports per-node meta data. Meta data in ARR is a list of key/value pairs that is bound to specific nodes in the graph. If you know the entity (e.g. through raycast), you can query its meta data through an asynchronous call as documented here: https://zcusa.951200.xyz/azure/remote-rendering/concepts/entities#querying-metadata
The meta data from the source file needs to be persisted during ARR conversion. This is explained briefly here: https://zcusa.951200.xyz/azure/remote-rendering/how-tos/conversion/configure-model-conversion#node-metadata
Let me know if you have further questions!
Florian