I solved a similar issue by building a knowledge graph. It took some manual curation and starting with a good base, but suggestions for misspelling and alternates were suggested by comparing vectors. The suggester runs as a batch with new entities after my ETL batch is done.