[wolfgarbe]

We are combining auto correction (SymSpell) and auto completion (PruningRadixTrie). That is sometimes tricky. We are using a static spelling correction dictionary which does not contains the term algolia, but angola. SymSpell is very fast, but requires a lot of memory. Therefore we are using a static dictionary, and not a dynamic dictionary derived from each individual index of each customer. Auto completion on the other hand we are dynamically generating for each individual index/customer.

[daibo]

I am combining symmetric deletes with a trie for fuzzy autocompletion and it is working pretty well.