About CaspiMind
CaspiMind turns equations into intuition.
We make visual explanations of the ideas behind math, computer science, AI, physics and signal processing. Every animation is built from a real simulation or a real measurement, so what you see is how the idea actually behaves, not a cartoon of it.
What we cover:
- Computer science: algorithms, data structures, and how CPUs, compilers and code really run
- AI & machine learning: tokenization, embeddings, transformers, optimization
- Math: calculus, linear algebra, probability, optimization
- Physics & signals: waves, Fourier transforms, filters, dynamics
Two formats: Shorts that explain one idea in a minute or two, and deep dives that build an idea from the ground up, with chapters.
Recent topics: why a sorted array is processed faster (branch prediction, re-measured on an Apple M4), how LLMs read text (tokenization), and why gradient descent gets stuck.
The name comes from the Caspian Sea, and the logo's three waves carry the colors of Azerbaijan, where the channel is made.
New videos on YouTube. Follow for the visual version of the ideas engineers use every day.
How the videos are made
Animations are written in Python with Manim. Every number on screen comes from a real run: optimizers on real loss landscapes, real tokenizer outputs, benchmarks measured on real hardware. Sources are credited on each video's page.