The demand for nuclear energy development is growing rapidly. As a fourth-generation advanced reactor, the small modular lead-cooled fast reactor is a highly promising development direction in the future. Our team has proposed a modular lead-cooled fast reactor model named “Lead Cube”, which has advantages such as “high safety, transportability, nuclear proliferation resistance, and low cost”, and can provide ideal solutions for energy supply to digital intelligence centers or remote areas. The reactor design of the “Lead Cube” involves the coupling of multiple systems and multi-physical fields, faces multiple condition constraints, and requires multi-objective optimization. Traditional reactor design methods are difficult to quickly and batch-compare and give a comprehensive optimal solution. In this work, a general design and optimization platform for reactors has been built. A “reverse training” method for BP neural networks has been proposed to solve the difficulty of specifying the enrichment of a scheme in advance. The genetic algorithm is used to search for optimal solutions. Finally, the rapid prediction of reactor parameters for the “Lead Cube” model and the multi-objective automatic optimization of batch schemes have been achieved. At the same time, this platform has good generalization ability and can be used for the design of other types of reactors.