Short description
Explore the Solar System at a true linear visual scale, with sourced guides and new learning journeys made by GPT-5.6 that happen inside the live simulation.
Inspiration
A lot of Solar System education teaches you the numbers but not what any of it actually looks like.
You get told how big the planets are, how far apart everything is, what their atmospheres are made of, and how they move. But most diagrams change the scale so much that you never really see those things together. You can know the facts without properly understanding how small, far apart, and different everything actually is.
ESS is for anyone who is curious about space, and also for educators who want a visual place to start. You can just open it and explore without having to begin a lesson or make an account. The educational parts are there when they can explain something you are already looking at.
What it does
EduSpaceSim shows body sizes and distances using the same linear visual mapping.
You can move around freely, select the Sun, every planet, or the Moon, and open a short sourced guide about it. There is also more detailed reading if you want it, but it does not force a wall of text in front of you every time you click something.
There is a short planner where you can choose suggested answers or write your own. GPT-5.6 uses those answers to make a new four-, five-, or six-stop journey from reviewed learning sections, supported camera views, and challenges.
The tour happens inside the actual simulation. It is not a separate slideshow or chat window.
The camera can approach a body, orbit around it, or follow a relationship between moving bodies. You can interrupt the camera whenever you want, look around yourself, return to the guided view, check the sources, find a body, ask for a hint, or leave the tour and go back to normal exploration.
There is also a reviewed fallback journey, so the experience still works when the live model request does not.
How it was built
ESS uses Next.js, React and Preact, Three.js, the OpenAI Responses API, GPT-5.6 Structured Outputs, Zod, Playwright, and Vitest.
GPT-5.6 does not write the scientific information shown in the experience. It can only return section, view, and challenge IDs from a list of things ESS already supports.
The server checks the response structure and whether the combination actually makes sense. If it does not, it gets rejected. ESS then loads the facts, sources, instructions, and camera actions locally from reviewed content.
So the model can decide how a journey is put together, but it cannot make up a new fact or randomly tell the simulation to do something it does not support.
The part that already existed before Build Week was an owner-controlled Solar System simulation taken from Actora.
During Build Week I made ESS into its own repository and deployment, removed private services and everything that was not related to the Solar System, added the scene-controller boundary, reviewed guides for the Sun, all eight planets, and the Moon, and built the planner, tours, challenges, GPT-5.6 integration, accessibility changes, and automated tests.
How Codex was used
Codex was my main engineering collaborator during the event.
It helped check and separate the useful parts of the existing Solar simulation, reproduce and fix camera and rendering problems, build the learning and model integration systems, add more tests, inspect the deployed judge experience, and keep track of the Build Week work and where everything came from.
I still directed what the product was supposed to be.
I decided to keep the scale realistic, only focus on the Solar System, make free exploration the main experience, use a neutral ESS guide, keep generated journeys limited to the current session, use reviewed information as the authority, and let people interrupt every guided camera shot.
Codex helped build it, but it did not decide what ESS should turn into.
Challenges and accomplishments
The camera work was probably the hardest part.
Following one planet is manageable. Following a moving relationship between multiple bodies while keeping the shot understandable is a different issue. The camera also had to deal with the visitor taking control and then returning to the tour without everything breaking or jumping somewhere strange.
Real scale creates another pretty obvious problem. Space is mostly empty and a lot of the bodies are tiny compared to the distances between them.
That is useful to show, because that is the entire point of using a real linear scale, but it can also make things hard to find. ESS had to keep the scale honest while still making small and distant targets possible to reach and look at.
The other challenge was deciding how much control GPT-5.6 should have.
I wanted it to make journeys that felt different depending on the person using it, but I did not want to treat whatever the model produced as scientific truth. The solution was to let it choose and combine reviewed parts instead of letting it generate the actual facts, sources, or simulation behavior.
The full experience now works without an account. You can start by exploring, answer a few questions, get a new tour, move through the cinematic views and challenges, see a recap, and then carry on exploring after it ends.
There is also a reviewed fallback and automated coverage for the main experience.
A normal planetarium shows you something, and a normal AI tutor talks about something. ESS connects the explanation, the camera view, the source, and the challenge to the same live scene.
What comes next
After Build Week I want to add a proper comparison tool, deeper guides for more moons, and more reviewed parts that GPT-5.6 can use when making tours.
I also want to improve the numerical scheduling before making claims about very precise positions or what the simulation shows over long periods of time.
Log in or sign up for Devpost to join the conversation.