AIstrophysics: launching LLMs into orbit
An LLM only becomes useful once it's connected to the right tools. A demo in orbit.
In this session, we'll start from a simple challenge: asking an LLM to model the solar system in 3D. Spoiler: it works… almost. The planets are there, so is the wow effect, but the textures are approximate, the positions questionable, and science quickly needs a bit more than “prompt magic”.
That's where things get interesting.
We'll enrich the LLM with open source data, introduce more realistic models and textures, then move to Python and more specialised libraries for more accurate results. Finally, we'll turn this experiment into a specialised agent, hosted in Microsoft Foundry as a Hosted Agent.
Through the fun lens of astrophysics, this session will show an essential reality for developers: LLMs are powerful, but they become truly useful when connected to APIs, tools, skills, domain libraries… and above all to humans able to check they don't drift off into space.
You'll leave with a concrete demonstration of how AI can augment entire fields, sometimes before those professions even realise it, and with ideas for applying these patterns to your own projects, your own users, and perhaps even your own job.
Our duo is a little unusual: even though we're both Program Managers at Microsoft, one is smarter than the other thanks to her background as an astrophysicist. A background that will come in very handy for validating the AI's work.
