Worked on enhancing compile-time performance in the espressif/llvm-project repository by implementing block-level side-effect analysis within MLIR. The primary contribution involved refactoring the wouldOpBeTriviallyDeadImpl function to process iterator ranges at the block level rather than per operation, allowing entire blocks to be queued for analysis and reducing redundant traversal of operations. This approach targeted performance bottlenecks in large programs, streamlining the analysis phase. The work leveraged C++ and focused on compiler development and performance optimization, demonstrating a deep understanding of MLIR internals and efficient code transformation techniques to improve the scalability of compile-time processes.