Mamori is a machine learning-based pathfinder for blockchain value extraction, focused on identifying zero-day economic exploits, MEV search, and intent resolution in smart contracts. Its core mechanism reduces all user-callable functions to state-changing functions, simulates blockchain state in a temporary in-memory database via stateful fuzzing, and employs particle swarm optimization to search for function call sequences that increase an attacker's balance, thereby uncovering economic security risks. It supports testing both deployed and pre-deployment contracts, and allows users to customize invariants and test priorities.