Learning in public, from the beginning
Zero to One AI is my record of learning artificial intelligence from first principles.
I am not trying to look like I already know everything. I want to preserve the more useful parts of learning: the wrong predictions, the experiments that changed my mind, the code that finally made an idea concrete, and the explanations that revealed what I still did not understand.
What belongs here
- Small, reproducible experiments
- Notes that connect equations to working code
- Comparisons with numbers, not just impressions
- Debugging stories and failed assumptions
- Clear explanations written after the idea finally clicks
The path currently runs from neural-network foundations through transformers, language models, evaluation, and practical AI systems.
The rule
Every post should leave behind at least one piece of evidence: a prediction, an experiment, a result, a failure, or a better mental model.
You can follow the work on GitHub or start with the learning roadmap.