True Knowledge is a natural-language question answering site that has the ability to infer answers to questions it has never been explicitly taught. It is built around a structured "knowledge-base" - a relational store of objects and "facts" that can be used by the inference engine to answer questions. The knowledge base is fed with a mixture of user-supplied facts and fact mined automatically when crawling the web. These are some brief designs for the consumer-facing corporate site. Click on the thumb-nails for full size images.
Essentially, the site is an interface to a user-sourced semantic store. Users "write" to the store by contributing either objects or facts (typed relations between objects), and "read" from the store by asking natural language questions. We have to make these complicated concepts effortless and even pleasurable.
To mirror the linguistic freedom when asking questions we also let the user enter facts in natural language. As long as we can recognise all the elements in his fact, and as long as the knowledge base contains an object for each element, we can enter the fact simply. When entering objects, we need a slightly more structured form - which the buttons along the bottom link to.
The answer page has to strike a balance between appealing to the casual user (who is only interested in getting his answer quickly), and the more technical user, or investor, who is interested in the technology involved. This design gives a lot of visual weight to the answer, but also allows the more curious user to disclose sections that reveal the steps our inference engine has gone through.
If we can't answer a visitors question we must be apologetic - the user must feel he is having a conversation real, polite person. We would also like him to fill the gap we've just discovered in our knowledge.