Worked on the e2b-dev/infra repository to deliver optimistic resource accounting for high-concurrency placement scenarios. Addressed the thundering herd problem by introducing active load prediction and immediate resource reservation, ensuring balanced placement even when node metrics were stale. Enhanced the backend API using Go, focusing on reliability and efficiency during placement storms. Developed and extended testing infrastructure with simulations such as NodeSimulator and LaggyNode, along with a load-visualization benchmark to validate distribution under challenging conditions. Prepared the codebase for controlled rollout by adding feature flag support, improving code hygiene, and refining shared packages to support maintainability and future enhancements.