Debtirtha Saha
Implementation writeups for neural networks built from the ground up. One project per post — the architecture, the bug that took longest to find, and the number at the end.
latest posts
| May 17, 2026 | Eight A100s, $61, and 124M parameters |
|---|---|
| Apr 25, 2026 | BPE from scratch, and why your LLM can't count L's |
| Apr 20, 2026 | Birkhoff in 8.7 KB |
| Apr 15, 2026 | Tiny Shakespeare, tiny GPT |
| Apr 08, 2026 | makemore: from counting bigrams to a WaveNet |
| Apr 01, 2026 | micrograd: a scalar-valued autograd engine |