SpenderQ-LYA — ML Pipeline for Large Quasar Spectra Catalogs

Public integration layer I built from the upstream SpenderQ framework (galactic-ai/SpenderQ) for the UW Bothell quasar research group: a data-driven Python stack that embeds the Spender convolutional spectrum autoencoder to reconstruct intrinsic quasar continua, mask absorption (including the Lyman-α forest), and batch-process large spectroscopic datasets. Complements the group’s EHVO analysis pipeline as an ML-based continuum fallback when classical normalization is insufficient.

• Built SpenderQ-LYA as a repository tuned for group workflows on high-volume SDSS-era quasar spectra, separate from the private EHVO branch.
• Implemented SpenderQ’s redshift-invariant latent encoding so continuum reconstruction scales across redshifts in our target range without hand-tuned templates.
• Uses iterative absorption masking so Ly-α forest structure can be separated from the underlying quasar spectrum in a simulation-free, data-driven way.
• Structured the repo for repeatable batch runs over thousands of spectra (the same scale as the outflow search pipeline).