A little less black box.
A working notebook about machine intelligence, written one idea at a time.
Latent Space is a place to slow down and understand how things work. The interesting part of a paper is often hiding between the architecture diagram and the implementation. That is the space this notebook explores.
What you will find here
- Machine learning foundations. The math behind the operations, with shapes and small examples.
- Computer vision. From patches and features to object detection and localization.
- Practical engineering. The memory, numerical, and implementation details that matter when theory becomes code.
The approach is simple: start with the intuition, draw the information flow, write down the equation, then build a small implementation you can inspect.
About this first edition
Start with the four-lesson foundations course , then continue to Attention & Transformers : eight connected lessons, from the first token vector to a small model you can train. Every equation has a term-by-term reading guide. Typed NumPy and PyTorch examples have a complete-source reader and recorded line-by-line walkthroughs. The final experiment includes a measured result on held-out sequences.
The course map also includes upcoming paths for advanced attention and for inference, deployment, and operating AI systems.
The remaining sample notes introduce computer vision and inference topics. The code is educational and deliberately compact. Each lesson or note links to primary papers or official documentation for deeper reading.
This page is ready for your own biography, research interests, and links. Replace it with the story of the person behind the notebook.
Follow along
You can subscribe with any feed reader using the RSS feed . New published notes will appear there automatically.