Celestine: AI Navigator for the Universe
Source: Dev.to
This is a submission for the DEV’s Worldwide Show and Tell Challenge Presented by Mux
What I Built
Celestine is an intelligent, multi‑modal AI navigator for the Solar System. It extends the intuitive experience of Google Maps to the cosmos, allowing users to explore planets in 3D and “land” on them to discover their secrets.
Unlike static star maps, Celestine features an AI co‑pilot (powered by Gemini 2.5) that acts as a bridge between alien worlds and our own. When you explore a crater on Mercury, the AI doesn’t just recite dry facts — it uses the Google Maps Platform to find a geological “twin” here on Earth (e.g., a similar crater in Arizona), instantly connecting the user’s cosmic journey to their home planet.
Gallery: Seeing the Universe
Here is a closer look at the key features of Celestine:
The Core Feature
The AI finds a “terrestrial twin” for a Venusian mountain range on Google Maps.
Multimodal Interaction
Users can talk to the AI via voice or a real‑time generated video avatar (powered by Tavus).
Back to Earth
The application visualizes the coordinates of earthly analogues on a 3D globe.
Planetary Navigation
The 3D interface allows seamless travel between celestial bodies.
My Pitch Video
(Insert video embed or link here)
Demo
Here is the code that powers the universe:
(Note: The project is containerized with Docker and deployed on Cloud Run, but currently requires a local setup for full interactivity due to cloud‑resource limits.)
The Story Behind It
Google Maps mastered navigation of Earth. But what about the rest of the Universe?
I built Celestine to revive the dream of a “Space Mode” — making it interactive, personal, and intelligent. I wanted a system where an AI agent could actually use tools — specifically the Google Maps Places API — to reason about geology and perform semantic searches across planets. This project is my first step toward making cosmic exploration deeply personal.
🏆 Recognition: This project was originally built for the Google Maps Platform Hackathon, where it was a nominee.
Official Showcase
Celestine featured as a nominee on the Google Maps Platform website.
You can view the original submission details here:
👉 View Original Devpost Submission
Technical Highlights
This is a full‑stack application that combines 3D rendering with advanced AI agent orchestration.
- Frontend: React +
react-three-fiberfor the immersive 3D solar system. - AI Engine: A multi‑agent system built with Google’s Agent Development Kit (ADK) and Gemini 2.5 Flash/Pro.
- The “Magic” Integration: The Analogues Specialist Agent analyzes celestial features and autonomously queries the Google Maps Places API to find terrestrial counterparts, returning coordinates that render dynamically on a 2D Earth map.
- Infrastructure: Backend in Python (FastAPI), containerized with Docker, and deployed on Google Cloud Run.
Architecture Diagram
Below is the high‑level architecture showing how the Multi‑Agent System orchestrates Gemini, Google Maps, and the frontend.
Challenges I Ran Into
Building a space AI is easier than teaching it to wait for Google Maps to load!
One of the biggest technical hurdles was handling race conditions where the UI tried to render a map before the API script was ready. I implemented a singleton loader pattern to solve this. Orchestrating the agents to handle voice, text, and visual data simultaneously required significant prompt engineering and logic design.
We are ready for takeoff! 🚀