My research lies at the intersection of AGN and galaxy evolution, observational cosmology, and astronomical data science. I use large, heterogeneous datasets from ground- and space-based surveys to study active galaxies, develop reliable methods for astronomical inference, and explore applications to cosmology and rare-object discovery.
A growing part of my work concerns science-driven machine learning for astronomy. I am interested in spectral representation learning, multimodal inference from imaging, spectroscopy, and catalogues, as well as similarity search and anomaly detection. The aim is to make astrophysical measurements more scalable without losing sight of uncertainty, survey systematics, or difficult cases that require human judgement.
My current research draws particularly on Euclid, Gaia, LOFAR, LAMOST, and multiwavelength archival data. I am also interested in future applications to DESI, CSST, Roman, Rubin/LSST, and other large survey datasets where the same questions of reliable inference and cross-survey analysis arise.
Large astronomical surveys, AGN & galaxy evolution
I study quasars and other active galactic nuclei as a population: how they are selected, how their spectra and multiwavelength properties vary, and what their hosts and environments tell us about black-hole and galaxy evolution. This work combines large catalogues with targeted spectroscopy and detailed analysis of individual sources.
The CatNorth catalogue and the CatSouth extension use Gaia astrometry together with optical and infrared surveys to construct reliable quasar samples across the sky. These catalogues support statistical AGN studies, spectroscopic target selection, and applications that require dense and well-characterised extragalactic reference samples.
Within the Euclid Consortium, I contribute to near-infrared data processing and AGN science. Our Euclid Q1 quasar study identified nearly 3500 quasars using NISP slitless spectra and examined their redshifts, colours, and spectral properties. Related work includes quasar host galaxies, radio galaxies and quasar environments in LOFAR data, and searches for unusual active galaxies.
My earlier survey of quasars behind the Galactic plane remains an important foundation of this programme. It combined machine-learning candidate selection with optical spectroscopy and later grew into a larger LAMOST sample. The project taught me how selection effects, heterogeneous photometry, and follow-up constraints shape a survey from the outset.
Astronomical data science & trustworthy AI
Large surveys need methods that can process millions of sources while retaining scientifically useful uncertainty and quality information. My work covers survey catalogue construction, spectroscopic redshift inference, representation learning, multimodal models, and tools for visual inspection and cross-survey analysis.
AIMS-z develops this direction for quasar redshifts. Its initial application is Euclid slitless spectroscopy, where sparse emission-line information and contamination can produce several plausible redshift solutions. The broader methodological questions include how to combine spectra with photometry and image information, transfer useful representations between surveys, detect domain shifts or missing modalities, and recognise when a prediction should not be trusted.
I am also interested in similarity retrieval and anomaly discovery as practical ways to explore survey archives. These methods are most useful when they help an astronomer move between population-level statistics and individual objects that deserve closer inspection.
Cosmological applications of large surveys
Quasars and galaxies trace large-scale structure over a wide range of redshifts. Building useful samples for cosmology requires accurate redshifts, understood selection functions, and careful treatment of spatial systematics. These requirements connect catalogue construction and astronomical data science directly to cosmological analysis.
I have contributed to work using the CatNorth quasar sample to investigate the S8 tension, alongside studies of radio-source environments and cross-survey source populations. These projects motivate further work on clustering, cross-correlations, and large-scale-structure measurements with Euclid, Gaia, DESI, LAMOST, and future survey data. This is a growing part of my research rather than a claim of a mature, standalone parameter-inference programme.
See Publications for papers across these themes.