When comparing technically complex and sometimes also quite abstract concepts at a detailed level (that's what patenting is all about!), traditional approaches reach their limits quickly. Keyword searches require long expertise and purely semantic searches are way too fuzzy. Both require huge amounts of time to reach decent confidence levels. Ever increasing amount of prior art does not make the problem any simpler.
Anyway, in their deepest essence, patents and inventions are not about terminology or semantics. It's all about technology. And technology is all about small elements interacting with each other to build something beyond the elements alone. In a sense, the definition of an invention is that 1 + 1 > 2. Building blocks of the invention produce something more than their obvious combination.
Traditional search approaches lack this very fundamental aspect and are therefore incapable of telling the computer what it should be looking for. IPRally is there to change that by introducing a formalistic model for technical data, the graphs. The graphs can express the technical relations and at the same time are sensible for the computer to process and understand. And equally importantly: easy for humans to understand too.
The graphs are a very expressive data construction, but still a simplification of full-text patent specifications and claims. Even though not all details can be preserved always, the ability to harness modern AI technology to do the dirty search work pays off quickly.