SPAMoE: Spectrum-Aware Hybrid Operator Framework for Full-Waveform Inversion

SPAMoE targets the intersection of AI4Science and Geophysics, specifically addressing the 'frequency entanglement' problem in Full-Waveform Inversion (FWI). FWI is a high-value, computationally expensive task used by energy companies to find oil, gas, and geothermal reservoirs. By applying Mixture-of-Experts (MoE) to different spectral components, the project offers a more nuanced approach than standard Fourier Neural Operators (FNO). The defensibility is currently low (4) because the project is in its infancy (9 days old, 0 stars) and exists primarily as an academic reference implementation. While it requires deep domain expertise to build, it lacks the 'data gravity' or community lock-in of established seismic packages like Madagascar or SeisSpace. Frontier labs (OpenAI/Google) are unlikely to compete here as the TAM is too niche and domain-specific. The primary threat comes from incumbent oil and gas service companies (Schlumberger/SLB, CGG) or specialized AI-for-Geophysics startups (e.g., Earth Science Analytics) adopting similar hybrid operator strategies. The 6 forks immediately following the paper's release suggest high interest among academic peers, but the project needs a more robust library structure to move beyond 'paperware' status.