KAIST announced on August 3 that a research team led by Professor Min-Soo Kim from the School of Computing developed RL-SPH (Reinforcement Learning-based Start Primal Heuristic), a reinforcement learning technique that trains AI to independently produce feasible plans without relying on an external solver. The technology learns how to produce solutions that satisfy multiple constraints encoded in an optimization problem, such as those in parcel delivery routing, vehicle routing, factory production scheduling, and hospital staff rostering, which can be formulated using integer linear programming (ILP). Existing learning-based approaches frequently violate constraints and often pass outputs to specialized ILP solvers like Gurobi or SCIP. RL-SPH addresses this by iteratively revising a candidate solution, selecting multiple decision variables likely to improve feasibility and determining whether their values should be increased, decreased, or left unchanged. The model learns from resulting changes in constraint violations and solution quality. The team designed the AI to first find a plan that is actually usable rather than the single best plan. The research team expects the method to serve as a foundation for AI-based decision-making in logistics, manufacturing, semiconductor production, and workforce management.