We ve all been there: you re designing a complex system architecture or a machine learning pipeline, and you hit a conceptual wall. You turn to Claude or GPT, but all they do is optimize inside your current, broken logic. They suggest refactoring loops or adding try-catch blocks, completely missing the fact that your entire problem representation is fundamentally flawed.
I believe the next generation of AI tools shouldn't just write mechanical code they must manage the topology of the problem itself, forcing "representation shifts" when localized search fails.
I m working on Research Space Navigator (RSN) to automate this mapping. We are launching in 29 days, but I want to build this out in public.
