The company I work for is running a project in which various numbers are getting scanned. Often, the barcodes were missing or illegible and had to be typed by hand. On the backend, we found that a great many of them were subtly wrong. For example, O (letter oh) and 0 (number zero) were swapped. Well, it's pretty easy to drop in a quick AJAX callback that checks the barcode number to make sure it is on file. I thought it would be cool, though, to have the program suggest the correct number to the user. If they were right and it was just something we hadn't seen yet, then they could just leave it be. If not, the system would give them a much better idea where they were messing up.
Meet the Levenshtein distance. I had heard of it before (it is commonly used in spellcheckers), but never had a reason to use it. A quick googling showed gave a blog post in which the writer implemented the dynamic programming implementation of an algorithm to find the Levenshtein distance as a MySQL UDF. It worked beautifully.