Software Engineer And Researcher
Current- Developing a novel method for isosurface extraction from implicit neural representations, focused on creating topologically consistent, non-overlapping, watertight, and visually appealing surfaces.
- Experimenting with Neural Network architectural features such as periodic activation functions, Fourier feature mappings, and custom loss functions incorporating geological data, such as ore grades.
- Revamped the Neural Network codebase by rewriting it in JAX and Keras3, which delivered a 5x speed-up in inference. This involved custom automatic differentiation and training routine implementations.