Speaker
Description
Large spectroscopic surveys provide an unprecedented foundation for astronomical discovery, yet scientific value does not emerge from data volume or model performance alone. Researchers must still reconcile heterogeneous data products, resolve scientific meaning, select appropriate analysis tools, assess uncertainty, and preserve the evidence required to reproduce a result.
This talk presents a service-oriented framework that turns multi-survey spectra into trustworthy scientific action. The framework begins with AI-ready spectral resources that preserve signal, uncertainty, quality flags, processing history, object identity, data versions, and provenance. An astronomical spectroscopy ontology then connects observations, physical quantities, celestial objects, analysis methods, and scientific evidence through explicit semantic constraints. Versioned scientific services provide validated capabilities for preprocessing, spectral measurement, parameter estimation, classification, redshift determination, and cross-survey retrieval.
Research agents operate on top of these resources and services. They interpret scientific questions, construct analysis plans, invoke appropriate tools, manage failures, compare competing hypotheses, and assemble evidence packages for expert review. Every result retains the data, ontology, model, and tool versions involved in its production.
Scientific cases are used as continuous system tests throughout requirements analysis, system design, development, integration, and acceptance. Examples include stellar parameter estimation, rare-object identification, galaxy redshift measurement, and multi-epoch analysis. Particular attention is given to rare objects and low-probability signals, which may be lost when knowledge systems rely recursively on model-generated content.
The goal is an open astronomical spectroscopy service in which data remain traceable, automated actions remain scientifically constrained, and every candidate discovery can be inspected, reproduced, challenged, and improved.