I have built a vector space model implementation as a Lucene-like search framework and I want to implement word2vec. In a setting where you at querying time have in your reach a char trie with addresses to posting files at the EndOfWord nodes and given that to solve this you can create any additional data structure and at indexing time store any additional data you need, how would you implement word2vec?

I have read a few articles about the theory but never read a word about implementing it. I could use a real walk-through here, with cheets and everything ;)