Inspiration
Mathematicians are used to constantly learning and exploring new concepts. One subject I am particularly passionate about is algebraic geometry. Roughly speaking, algebraic geometry studies the geometry of equations, such as $$x^2+y^2=1,$$ using algebraic methods. However, to have access to essential characteristics one needs sophisticated definitions and constructions. This is where the Stacks Project comes in. It is a collaborative, open-source reference that presents these concepts and constructions in remarkable detail, spanning more than 7,654 pages, 3,298 sections, and 116 chapters! The question, then, is how to engage with this incredible resource. Reading it directly from beginning to end, line by line, does not seem like the best approach. In fact, the Stacks Project itself does not recommend doing so. Our website proposes a different approach: transforming this vast resource into an “Exploration Mode.” We hope that the project can also serve as a model for developing new and more exploratory ways of learning powered by AI.
What it does
Our project has two main parts. The first processes files from the Stacks Project and uses its tagging and indexing system to create the skeleton of an example-driven interactive experience. It also generates a dependency graph connecting the different topics, which serves as a map for exploration. The second part takes this processed data and turns it into a website that guides the user through the material.
The website consists of a graph view and a conversation panel. Users can click on any node in the graph to begin exploring its content through a structured conversation. Their progress is tracked using simple visual indicators. All conversations are designed to encourage an exploratory approach to learning. The website combines the graph and the parsed mathematical content to create an exploration-driven experience: the graph serves as the map, while the conversations guide the learning process.
Finally, the website includes two characters that help guide the user. The first is Shurik, a nickname used by Alexander Grothendieck, one of the principal architects of modern algebraic geometry. Shurik acts as a guide through the content of the Stacks Project and refers only to information obtained directly from it. The second character is the Oracle, an out-of this-world, computer-like being with mysterious knowledge. The Oracle offers indirect advice, hints, and additional perspectives. It represents the AI-powered aspect of the website.
How we built it
We built most of the project using ChatGPT 5.6 Pro. The code is relatively simple and transparent, to the point that even someone without extensive programming experience, like me, can understand and follow its general structure.
Our basic approach was to parse the Stacks Project files, extract their tags and dependencies, and use this information to generate both the exploration map and the interactive conversations.
Challenges we ran into
One of the most challenging, and exciting, parts was developing the idea of the Oracle. We also had to carefully tune the conversations so that they felt as exploratory and engaging as possible. Another challenge was finding the right balance between guiding the user and allowing them to discover mathematical ideas independently.
Accomplishments that we're proud of
The entire process gave me a new way to express and develop mathematical ideas in this medium. Although I had previously created academic websites for myself and for conferences, I had never built anything of this scope. More importantly, the project showed me how much more there is to explore and create. The current website feels not only like a finished prototype, but also like the beginning of a much larger project.
What we learned
We learned many things during this week. Before this project, I had no experience building this type of interactive website and, to be fair, very little experience with coding. I learned how mathematical content can be transformed into an interactive structure, how dependency graphs can support navigation, and how AI can help create a more exploratory learning experience. I also learned that it is possible to build a substantial project while learning many of the necessary technical skills along the way.
What's next for Exploration Mode: Stacks Project
There are many possible directions for this project. Let me highlight the ones that I find most exciting.
During the conversations, users are encouraged to think of new examples, counterexamples, and possible variations of the hypotheses. I would therefore like to develop an “Explorer Notebook” associated with each conversation. It would allow users to record their examples, questions, observations, and discoveries as they explore the material. Later, when these can be compiled with all the references they are associated.
Another goal is to improve the Oracle. It could provide more examples and context during conversations, use the user’s interactions to personalize its suggestions, and recommend further reading outside the Stacks Project. For instance, it could suggest arXiv papers in which the topic currently being explored is applied.
A further direction would be to begin with a collection of lecture notes written in LaTeX and automatically create a website similar to the Stacks Project experience. Our “Exploration Mode” could then be applied to the resulting set of tags and dependencies, creating a customized learning environment for the students in that particular course.
Log in or sign up for Devpost to join the conversation.