> Provide real-time text search over a large corpus (ie, some subset of Project Gutenburg, a bunch of product reviews, etc.) That would be nice. But have you looked at the PG export yet? It's a million of individual files in the semantic web notation with duplication for the author information, etc. I would LOVE for somebody to convert that into the full-fidelity form suitable for import into the search engine. So, please do it for real, not just as a 'possible'.

Another dataset with similar problem is Google Mail takeout. Theoretically in an mbox form, there is apparently enough quirks in there to not be parsable by the standalone 3rd party libraries. Somebody said python might be able to do it, but I haven't seen the confirmation specifically for Google Mail.

If you manage to do either and document the steps/share the code, I'll personally sing your praises at the search engine workshops/meetup/lectures I do (usually Solr rather ES).