Rushd: A Multi-Specialist Reasoning System with Learned Router on Consumer Hardware

Large language models (LLMs) have demonstrated remarkable capabilities across a broad spectrum of reasoning tasks, from mathematical problem-solving to strategic planning. However, monolithic models — even those with hundreds of billions of parameters — face fundamental trade-offs between generality and specialization. A single model tasked with logical deduction, planning, decision-making, and game-theoretic analysis must distribute its limited representational capacity across all these domains, potentially sacrificing depth in any given one.