Tutorials show one move at a time, cleanly, with the mistakes edited out. This is the other kind of video: a screen recording of a real session, start to finish, played at double speed so that eleven minutes of work fit in five and a half. There is no narration and nothing was cut. Music, a mouse, and a graph growing.
The subject is stochastic calculus, chosen because it is dense with equations and because its concepts relate to one another in several distinct ways: Brownian motion is a foundation, Itô’s Lemma is a result, the Itô correction term is a part of that result. A graph that gets those relationships right looks different from a list of the same names.
Building a knowledge graph in real time with Knogra. Five and a half minutes at double speed; no narration.
What happens
The workspace starts empty. The first thing typed into the chat is not a command but a question: “I want to study Stochastic Calculus. Can you advise me where to start?” The assistant answers as a tutor would, with a short overview of the subject and a suggested order of study, and alongside the prose it proposes the first handful of concepts. Those proposals arrive on the node shelf, each with its equation already written, and are accepted one by one onto the canvas.
From there the session moves the way real work moves. Nodes get dragged into a layout that reads well. The view steps into Brownian Motion, then into Itô’s Lemma, and each step is a scene change: the new central node takes the middle and the graph rearranges itself around it. Some requests to the assistant are made from the menu, some are typed.
The request worth watching for is the one made on Itô’s Lemma: not “suggest related concepts” but “suggest children, of type Composition”, which asks for the parts of the lemma and nothing else. The assistant returns the Itô correction term, the diffusion term, and the multiplication table, each connected as a component. Then a typed follow-up, “and what about drift term? deterministic term I mean, shall I have a node for that one too?”, and the assistant agrees, explains why, and proposes the missing piece.
That exchange is the point of the video. The assistant did not decide what the graph contains; it proposed, the user questioned, and the graph ended up holding what the user judged it should. Every node on screen was suggested by the model and every one was accepted by a person.
Things to notice
Equations arrive with the nodes. The assistant writes each concept’s canonical formula as part of the proposal, and the canvas renders it. Nothing was typed in LaTeX by hand.
The relationship types are not all the same. Edges from Stochastic Calculus to its main topics, edges from Itô’s Lemma to its terms, and the edge between the correction term and the diffusion term are different kinds of connection, and the assistant was asked for them by kind. How that works, and why it matters, is the subject of a separate post on relationship types.
The graph is never finished. The video stops at a convenient moment, not at a conclusion, which is what building a graph of a real subject looks like.
If you want to try the same thing, open app.knogra.io, start a new graph, and type a question about something you want to learn. The basics series covers the individual moves if any of them are unfamiliar.