Grid provides efficient environment to execute application faster, with desired Quality of Service (QoS) constraints. Performance of grid (mainly time to execute applications) is dependent on the job scheduling strategy used to map applications or the job (collection of atomic tasks) grid resources. Grid Scheduling Algorithms (GSA) generates schedule of jobs and the corresponding resource which satisfy requirement of the job for successful execution. Attempts in optimizing the schedule shows use of classical, heuristic and nature based algorithms. For instance Genetic Algorithm (GA), Simulated Annealing (SA), Ant Colony Optimization (ACO) and Neural Network and Fuzzy System (NNFS) prove to be more efficient than classical scheduling algorithms. The performance of grid using discussed GSA requires improvement which will result in comparatively optimized schedule. The scheduling process needs to be more dynamic so as to consider runtime changes in the resource properties (for example CPU time, Memory, etc.) We implement ACO and GA as separate GSA used for the design of our scheduler of our experimental grid. We propose multilevel decision making engine, as special case Two Level Decision Algorithm (TLDA). In the first level decisions we generate initial schedule using Ant Colony Optimization algorithm (ACO). We refine this schedule using Genetic Algorithm (GA). This paper presents the findings of our proposed approach of multilevel decision engine as compared to the performance of GA, ACO each. Results show considerable improvement in execution time than the former approaches for scheduling.