ARouter

This project addresses the critical challenge of deploying highly accurate AI solutions within heavily constrained hardware environments. At its core, the system acts as an intelligent gateway that evaluates incoming tasks in real-time, determining the optimal execution path without exceeding strict memory or processing limits. By implementing a dynamic evaluation layer, the architecture seamlessly shifts between lightweight, edge-optimized processing and powerful remote APIs. This ensures that straightforward deterministic tasks are handled locally to minimize external API costs, while complex logical, mathematical, or coding challenges are automatically escalated to heavier models. To guarantee stability under load, the system employs aggressive resource-capping mechanisms, preventing the memory overruns that typically cause container crashes in restrictive environments. The result is a highly resilient, cost-effective pipeline that maximizes overall benchmark accuracy, gracefully handling diverse data inputs while maintaining absolute stability from cold start to final output.