Worked on the adap/flower repository to enhance reliability and resource management in distributed environments using Python. Addressed a critical issue in Ray integration by implementing robust logic in pool_size_from_resources, ensuring that nodes reporting no CPU or GPU resources are handled gracefully. This fix prevents mis-sized resource pools and allocation errors in heterogeneous clusters, directly improving system stability. Automated tests were added to validate correct behavior in mixed-resource scenarios, supporting ongoing maintainability and early detection of regressions. The work leveraged backend development and distributed systems expertise, resulting in more predictable and resilient resource allocation across production deployments in adap/flower.