You might actually find Dogpile (also by Mike Bayer) more approachable and useful in this context. Even the separation of the project into logically distinct packages is instructive.
This one could be very interesting as I have decades of experience with DBs and SQL, so I was always wondering about ORMs, and sqlalchemy seems to be best in field at the moment.
Large pieces of SQLAlchemy are about metaprogramming and some syntactic magic required for an easy-to-use ORM. It's mostly nice on the inside (to my cursory glance), but neither simple nor easy for an inexperienced person.
Still it's a good example how to make many relatively complex parts easily composable and efficient.
- https://github.com/zzzeek/sqlalchemy
- http://www.aosabook.org/en/sqlalchemy.html
- http://docs.sqlalchemy.org/en/rel_1_0/