I'm wondering a little about the accuracy of the searches, particularly in regard to differentiating web searching from data mining. Granted, the context in which the search tool will be used likely shaped the principle methods by which searches are completed. If the software "runs" like a typical search engine, it may be prone to a common problem with search-enabled collections of data: Even after optimization (within the search utility and modifications to the data), search algorithms break down as the size of the data collection grows. I'd also like to explore how the tool accounts for the relationship among search terms or phrases. Is there an accommodation for moving from broad search phrases to more narrow and specific terms (as most of us have been trained to do using common web search tools)?
Again, there is obvious application here for improving online learning experiences. The possibility of incorporating a knowledge path that moves from the textual elements of a course to a dynamic, voice-enabled search interface, which then presents learner-selected information aurally and visually... moving back from the lecture to a synchronous environment in which learners discuss their presentations, verbally and textually.