Worked on the lattice/quda repository to enhance the IRAM eigensolver by introducing a configurable QR tolerance parameter, aimed at improving numerical stability and user control in large-scale simulations. The implementation added a new qr_tol setting, with its default value dynamically derived from the main eigenvalue tolerance, and included warnings for potentially unstable configurations. Adjustments to the default and warning thresholds reduced the risk of unstable convergence, minimizing the need for manual tuning by researchers. This feature was developed using C++ and leveraged expertise in high-performance computing, linear algebra, and numerical methods to strengthen the robustness of eigenvalue computations.