This library follows a schema-less approach, but I was wondering whether a schema-based option is not worth considering.
Adding a schema in the mix could have a couple of advantages:

• a more optimal conversion that leads to better performance (queries are predictable)
• clarity for client developers by supporting graphiql, etc.
• compatibility with other graphql tools;
• it's complimentary; supporting a graphql schema does not prevent the current approach
• no need to start from scratch, there are plenty of examples out there in other programming languages.

(this could be something we can put a bounty on)