A new approach to seismic data reconstruction called SeisEvo uses a multi-agent search driven by large language models (LLMs) to evolve algorithms that can be used for tasks such as interpolation and denoising. Unlike traditional methods, which rely on manually designed structural priors and iterative operators, SeisEvo searches for the algorithm that produces a specific reconstruction result. The approach is able to discover new operators that outperform existing ones, including a residual-gated, phase-aligned dip-consistency projection and a reliability-grouped singular-value shrinkage. These operators can be used as standalone white-box algorithms without requiring an agent or neural network at inference time. --- Why it matters: This matters because it shows how agentic algorithm evolution can complement deep learning in discovering explicit, inspectable, and deployable seismic processing algorithms, which could improve the accuracy of seismic data reconstruction. Source: https://arxiv.org/abs/2608.18272