Introduction

Welcome to the documentation for the Lasair Virtual Research Assistants (VRAs), machine learning algorithms that will provide further annotation to the LSST stream on the Lasair platform.

I am an extra-galactic transient astronomer looking for supernovae and other distant explosions.

Important

These algorithms are trained directly on the live Rubin stream, using human-made labels provided via eyeballing. Because I am the eyeballer, my scientific biases can affect the performance of the tools I build. If you are an AGN, CV or Solar System Person, take what I do with a pinch of salt. I will always welcome your feedback as well, feel free to email me.

What is a “Virtual Research Assistant”?

They’re just my bots that I train myself on bespoke astronomical data; they are not Agentic AI. If you want to see the work I did for the ATLAS Virtual Research Assistant, see Stevance et al. 2025

Note

My methods use Feature based ML, exclusively trained on real data, and as of recently focus on using Active Learning methods so I can train directly on Rubin as a sole eyeballer (they allow me to train algos with a few hundred examples instead of thousands or millions).

If you see something tagged VRA or LVRA, it’s probably something I made and if you need someone to blame it will likely be me. I give them human-like names with l33t-like formatting, which is another way to recognise them.

The first Lasair VRA is r0b, my real-or-bogus classifier to help me find real extra-galactic events. More docs to in the next few weeks.