Research Ecosystem
A map of how the three projects connect, what they share, and which scientific questions each one is built to answer.
Research program map
Three projects, one coherent scientific direction
The ecosystem connects stellar spectroscopy, galaxy morphology, and Bayesian cosmology through a shared philosophy: physics-grounded, interpretable, reproducible machine learning and inference for astronomy.
Connection map
How the projects connect
The ecosystem is not a loose collection of repos. It is a structured set of research programs connected by questions, methods, and scientific constraints.
Dataset / input
LAMOST DR5 spectra
Research program
AstroSpectro
Scientific outcome
Stellar classification, feature engineering, SHAP, and latent-structure analysis
Dataset / input
Galaxy10 DECaLS
Research program
AstroVision
Scientific outcome
Morphological classification, representation learning, multimodal fusion, and physical structure analysis
Dataset / input
Pantheon+SH0ES + DESI
Research program
ξ Dark Energy
Scientific outcome
Bayesian model comparison, parameter inference, diagnostics, and explicit cosmological caveats
Shared themes
Common themes across projects
These themes are the connective tissue of the ecosystem. They explain why the projects belong together even when the data, methods, and scales are different.
Interpretability
Each project tries to open the black box, but in a project-specific way.
Uncertainty
Predictions and conclusions matter only if uncertainty is made visible and discussable.
Reproducibility
Every branch of the ecosystem is designed to leave behind a traceable research path.
Physical grounding
The final criterion is not only performance, but scientific meaning.
Scientific questions
What each project is really asking
These are not just engineering objectives. Each project is anchored in a scientific question that gives the technical work its meaning.
What physical information do modern machine-learning models actually learn from stellar spectra?
The project is built around the idea that interpretability only matters scientifically if learned structure can be tied back to astrophysical signal.
Open overviewCan self-supervised visual representations trained outside astronomy capture physically meaningful galaxy structure?
AstroVision tests whether modern visual encoders can become more than classifiers: tools for representation, morphology, and scientific interpretation.
Open overviewHow do alternative dark-energy parameterisations behave under current observational constraints?
The cosmology branch explores model behaviour carefully, with transparency about what is constrained, what remains ambiguous, and what stays exploratory.
Open overviewUnifying principles
What really holds the ecosystem together
These principles are the closest thing to a common research thesis across the three programs.
Scientific ML, not benchmark chasing
The ecosystem values scientific understanding, physical relevance, and interpretability more than leaderboard-style optimisation.
Reproducibility as a research requirement
Pipelines, notebooks, and decisions should remain intelligible, rerunnable, and useful to a future collaborator — or to future Alex.
Physical meaning over black-box convenience
Methods are selected not only because they work, but because they can be argued for scientifically and linked back to the problem domain.
Cross-project transfer of methods
The three projects are distinct, but they cross-pollinate through shared themes: interpretability, uncertainty, representation, and Bayesian reasoning.
Continue exploring
Choose your next entry point
Once the ecosystem map is clear, the best next step is to move toward a specific project or reading mode.
Open AstroSpectro
Start with the flagship spectroscopy program and its scientific documentation.
Go thereRead the journal
See how the ecosystem evolves over time through milestones, notes, and design decisions.
Go thereFollow a reading path
Use guided entry points based on what you want to learn or reproduce.
Go there