A Python Runner to Extend Seismic Processing in the AI Era

Seismic processing geophysicists draw business value from raw data through carefully tuned transformations, from denoise to demultiple and FWI. The modern Python ecosystem brings new tools to this kind of work. Libraries such as PyTorch, JAX, and Transformers deliver high-performance computation by wrapping optimized native code in concise Python interfaces. Yet isolating and deploying these dependency stacks inside enterprise pipelines poses challenges.

We present a Python runner designed to address these challenges for seismic pipelines. We target two design principles: (1) build with the ecosystem, not against it, and (2) provide developers a clear path to scale.

The runner framework defines a lightweight in-memory data layer on Apache Arrow. Arrow Flight gives process separation, so user dependencies do not collide with platform dependencies. With this, geophysicists and developers can write ML and AI applications in Python and run them directly in processing workflows.