Building an index of suggestions is indeed something we’re exploring at Algolia, but what we’re building requires quite a lot of traffic for it to make it relevant: that’s why it’s currently only for some of our Pro or Enterprise customers.
That being said, there are ways to achieve what you’re doing - but it requires a handful of search operations. Basically, you could run a search for every “Top Searches”, and look at the highlighted results for the first few results and see if it matches a full word.
In your example, if you were to run search with the query “refr”, you would see that the results would all match the word “refrigerator” (in the _highlightedResult part of the response), and thus would make you confident that “refr” is only a prefix match.
However, be careful when building suggestions on top of your Top Searches. If your users figure it out, you could be a victim of spam or cheat. See how it happened for a large French retailer last year: https://nakedsecurity.sophos.com/2016/06/10/diy-stores-poisoned-search-engine-returns-nasty-results/
Hope this is helpful.