Hello.
Here in Brazil we are in early stage of developing a blend4web based system to search, present and checkout sign language animations. The following image represents the system.



The animations are motion captured with a Vicon system and postprocessed in Blender using Makehuman characters. Since it uses motion capture, each action is heavily keyframed. Individual signs are size compatible for loading the JSON. But we are dealing with a database of over 6000 actions.
Is there a way to store actions in a separate .blend file (or files) in the server and load/append them dinamically in the main JSON?

A very unoptimized version with the avatar performing some actions may be checked here (content only in portuguses so far).

Thank in advance;

Ângelo Benetti
Motion Capture and Blender specialist
CTI Renato Archer - MCTIC - Brazil