SkyMiner AI Ecosystem

What is SkyMiner AI?

Modern astronomical surveys generate more data than any research team can manually review. DESI has catalogued 9 million spectra of galaxies, quasars, and stars. The Zwicky Transient Facility measures how nearly 5 billion objects change in brightness over time. Somewhere in those numbers are objects that behave in ways science has not yet documented.

SkyMiner AI is an independent research project trying to find them — without being told what to look for.

The Approach

Every pipeline in SkyMiner operates without labeled training data. There are no annotated examples of “this is an anomaly” — because genuinely novel objects have no precedent to learn from. Instead, each pipeline learns the structure of normal, and scores how far a candidate deviates from it. All parameters are derived from the data, not assumed.

The architectural decisions involved — which representations to use, how to handle irregular time series, how to define novelty without ground truth — are problems without off-the-shelf answers. This project is, in part, a demonstration of applied AI at a scale and depth that most teams approach with considerably more resources.

Pipelines

SkyMiner Spectral

DESI DR1 — 9.1M spectra

Looks for any object — galaxy, quasar, star, or otherwise — whose optical spectrum doesn't fit a known category. Processes 9.1M DESI spectra through a custom neural encoder and flags what sits outside the learned distribution: unusual emission lines, unexpected continua, features that match no standard template.

IN PROGRESS

SkyMiner Temporal

ZTF DR24 — 4.97B lightcurves

Looks for stars and transients with unusual variability patterns. A custom Transformer trained on ZTF lightcurves — without labeled examples — learns what normal variability looks like, and flags what doesn't fit.

DESIGN COMPLETE

SkyMiner Fusion

DESI + ZTF — multi-domain synthesis

When an object appears anomalous in both its spectrum and its brightness history, independently, the case for novelty is much stronger. Fusion cross-matches signals from Spectral and Temporal to identify these compound candidates.

BLOCKED — Spectral + Temporal required

SkyMiner Orbital

ZTF DR24 — Solar System Objects

Looks for Solar System objects whose orbital parameters don't match anything in the Minor Planet Center catalog. Operates entirely independently of the other three pipelines.

IDEATION

Pipeline Dependencies

SkyMiner Fusion requires both Spectral and Temporal complete before it can run. SkyMiner Orbital is fully independent and can proceed in parallel with the other three.

Ecosystem Status

Status Legend

About

SkyMiner AI is an independent project by Hugo Zanqueta — a systems engineering leader with 20+ years in multi-physics product development, published researcher in acoustic engineering, and founder of AlignIQ Advisory. Across domains, the consistent pattern: a refusal to stop at the surface-level answer, and a drive to turn genuine technical curiosity into innovation that holds under scrutiny.

The same methodology that defines that work — traceable parameters, empirical validation, no assumption without evidence — is what drives SkyMiner. A different domain, built with the same rigor. The goal is to find objects that existing catalogs, conventional pipelines, and standard survey processing have missed. The work is ongoing.

Contact

Questions, collaboration ideas, or just want to discuss the project? Send a message below.

Alternatively, reach directly via LinkedIn.