Feedback on text recommendation engine

For a project I was working on, I made a proof-of-concept text recommendation engine

However, my project got cancelled. That said, I had a lot of fun building it and would like to continue to work on it.

• Is something like that any useful to anyone else?
• If so, in what disciplines would something like that be useful?

The basic idea is that documents that have similar "semantic" topic distributions would be similar.

Here's how it works:

• I have implemented an application that will use Latent Dirichlet Allocation to create topic models from a given corpus of data
• Subsequent document queries will have their topic distribution inferred and matched for similarity via Random Projections and Levenshtein distance

You can find an example running here: http://174.129.217.121/smarts/query/hnlinks/

If you go to hacker news, and pick a url for an article and enter it in the textbox, there is a chance similar content will show up.

EDIT: formatting

EDIT2: The training set hasn't been picked for accuracy, so the recommendation results might be inaccurate unless its articles about education, startup fanboy pieces, nerd-shoe-gaze blog posts and articles about apple and google.

EDIT3: The demo server now runs on port 80!