At the moment, the algorithm utilizes various tasting notes and other characteristics (price, etc) to determine what is the closest fit. While this has much room to improve, the general idea is there. I would like to get it to a point where users can create their own profiles with their personal ratings and then use that data to make the suggestions more accurate.

This is exactly what I was thinking, I was trying to avoid suggestions that end up being a $10k bottle of whisky, I wanted them to be as similar as possible, without needing to ask the user too many questions.

Once I start getting into the big crowdsourced data sets is when this will get really interesting to see the connections between preferences.

Sounds like I need to try this Eagle Rare! I'm sure you've tried some of these already, but my personal recommendations are Lagavulin, Glengoyne and Edradour.