AstroLLM Assistant: Astrophotography Meets Gen-AI Large Language Models (LLM)

Introduction

Recently I started to take astrophotographs of solar system and deep space (moon, galaxies, nebulae, star clusters) that adorn our night sky. I call these photos astro-art-over-home. As far as astrophotography is concerned I am just an amateur and rookie. I take those photos with a so-called smart telescope (good tool for rookie learners). While publishing these photos I decided to create short astronomy knowledge captions for the photos that include information on how far they are from earth in light-years, why they look the way they do like nebula colors, etc., and a little bit of astrophysics related information about those celestial objects. Not being an astronomy and astrophysics expert I had to collect the caption information somehow, hopefully from reputed sources. Manual curation of online sources could have been a way to go. But I decided to automate the collection process utilizing Gen-AI LLM (large Language Model) assistant and develop the assistant on my own. So along with astrophotography hobby I embarked on building a (personal) system I call AstroLLM Assistant.

AstroLLM System and LLM Topics

The AstroLLM RAG system and relevant LLM topics, including Multi-modal/Vision models and LLM hallucinations are described in the AstroLLM System page.

Astro-Graph

The Astro-Graph (generated by the AstroLLM system) is a fun way of showing my celestial photos in the context of Constellation and distances in light-years (the lengths of the edges/lines between Astro-Graph nodes have been pseudo-scaled). The objects are clustered in respective constellation box. With clustering the distance represented with pseudo edge scaling is lost. For example, when we take photo of the West Veil Nebula we see a bright star (52 Cygni) in the middle of the photo giving an impression of it being part of that nebula. But the star is pretty far away from the nebula. One of the purposes of the Astro-Graph is to highlight that. The Astro-Graph is auto-generated based on distance information output by Astro-LLM Assistant, which is then processed to output graph with Graphviz (AstroLLM can be instructed to output graph structure in a graph description language. See an example here where advanced prompt engineering was used to instruct LLM to generate graph definition and then BGP network topology graph was rendered).

An abstract /fun way of looking at the Astro-Graph is as if astro arts (based on celestial photos I snap) are hanging over my (yours too) home!

Astro-Map

The astronomy map of some of my astrophotographs based on Hubble Skymap is shown below. The Milky Way image (which is not a real photo) in the map is based on ESA (European Space Agency) Gaia star map.

Astrophotographs

Photos of moon, galaxies, nebulae and star clusters were taken with exposure times of 30 minutes to two hours. I am a rookie astrophotographer. The post-processing of the photos was rudimentary (minor touches with Mac Photo app).

Images (except GAIA image), system/development architecture are personal

Astronomy information sources (for RAG): Wiki, NASA, ESA

Title Image: Dumbbell Nebula snapped by MZH

Page updated

Google Sites

Report abuse