Developer builds transformer LLM from scratch in TypeScript

By Splitfeed EditorialLast updated Oct 14, 2026, 10:40 UTC

On October 14, 2026, a developer posted to Hacker News documenting an ongoing project to build a ChatGPT-style transformer large language model from scratch using TypeScript. The project is structured in chapters, with completed sections covering dataloaders, embeddings, and the attention mechanism. The author is actively soliciting feedback and the project remains unfinished.

The deliberate choice of TypeScript over Python is the most substantive editorial decision here. Python dominates machine learning tooling, and most from-scratch tutorials, including influential ones like Andrej Karpathy's, use it. The author's argument is that working in a less familiar language strips away the comfort of well-worn libraries and forces an encounter with implementation details that Python's ecosystem can obscure. Whether that pedagogical theory holds is genuinely open to debate, and the project's ultimate value will depend partly on how far it gets and whether the TypeScript implementation surfaces any real surprises compared to standard Python walkthroughs.

No reactions from the Pro-AI, Anti-AI, or Middle Ground camps have been published yet. Pro-AI voices would typically celebrate a project like this as evidence that AI literacy is spreading and that the tools and concepts behind large models are becoming accessible to a broader range of developers. Anti-AI commentators would more likely focus on what the proliferation of such tutorials means for the pace of AI development and who ends up building with that knowledge. Middle Ground observers would probably treat it as straightforwardly educational, worth encouraging while noting that understanding a transformer architecture does not resolve the harder questions about how these systems are deployed and governed.

The natural thing to watch is whether the author completes the remaining chapters, particularly the training loop and any inference implementation, which would represent the more technically demanding and revealing parts of the build. Community feedback on Hacker News in the days following the October 14 post may also surface concrete critiques of the TypeScript approach that would test the author's core argument.