A document search index with the following goals:
- High performance type-ahead search for 1M documents, each containing a few hundred words.
- Persistence to disk.
TODOs: [x] Use spelling correction for queries to make interaction easy. [x] Run simple HTTP JSON API server. [x] Build simple HTML and JS testing interface. [ ] Load large genealogy data set into index and interact with it in the UI. [ ] Fix "single character" query term issue. Shouldn't have empty spelling correction suggestions. [ ] Sort search results [ ] TF-IDF? How closely words appear together? Prioritize field? [ ] Persist data to disk to recover quickly from closing, crashes.