Mapping indexes to look-up tables instead of secondary indexes in Cassandra may be a good first step on the path to the denormalization engine.

The performance of secondary indexes in Cassandra highly depends on the cardinality of indexed fields, as e.g. described in this blog post. So it may be beneficial in many cases to create a separate look-up table keyed by the field of interest.

This would also allow to map multi-column indexes if the index fields allow to create a combined key representation.