My initial thoughts here will be that we can provide node library that will include all execution logic for graphql and our custom resolvers format. UI then can use this to mount their own graphql logic and test it.
Having it this way will allow us to separate development UI and production (server) and allow to stage and test changes in UI and then "publish" them to server. This aproach has many benefits, however separate library introduces additional complexity - but we can avoid that by having one general datasync repository with layout:

- UI

- Framework

- Server

- Example data (currently sequelize)

Having it this way will be much better for local usage where docker-compose could be used to run all of this, but it will make CircleCI config more complex.