As I got more and more involved in the Starknet ecosystem, discussion channels started flooding with questions that could simply have been answered by linking the right resource. At the time (early 2024), LLMs were getting good enough to be helpful chatbots, and I started thinking about how to architect a RAG-based system that would answer most of those questions and point people in the right direction.
I built Starknet Agent in collaboration with the Starknet Foundation, owning the entire system end-to-end, running evaluations, and using it myself for my day-to-day work.