- Better handle uploaded fulltext files (really, all data artifacts saved on disk, not in the db)
- Allow for assigning studies to reviewers for screening?
- Add extra fields in users table, e.g. affiliation?
- Improve and extend NLP functionality, just across the board
- Enable https everywhere (via let's encrypt?)
- Consider removing async tasks (and db columns?) for text content vectors
- Add script to train a v2 (splink-based) deduper model
- Add notebook demonstrating functionality of various NLP models
- Only clear minimal subset of authz caches when review status/membership changes