This paper describes the design of a new ScaleText system aimed at scalable semantic indexing of heterogeneous textual corpora. We discuss the design decisions that lead to a modular system architecture for indexing and searching using semantic vectors of document segments – nuggets of wisdom. The prototype system implementation is evaluated by applying Latent Semantic Indexing (LSI) on the Enron corpus. And the Bpref measure is used to automate comparing the performance of different algorithms and system configurations.