Yeah thats the main differentiation - our backgrounds are in search, and we think its pretty hard to use traditional chunk -> embed -> retrieve- > rerank for structured data (they work well for unstructured data).

So we use a different approach where we create the code graph and apply classic ML (like clustering) to find abstracted patterns in the graph. We have AI agents that traverse these abstractions to create knowledge which we search on.

This enables better quality responses for more complex questions about generic higher level concepts.

We think of it in the following way: Cursor etc will be extensively used to write code. Our tool comes into play when you don't know exactly where to start, which is why we focus on KPIs like new eng onboarding times, reduced eng hours on KT, etc - overall developer efficiency. Lmk if it makes sense!