[{"content":"I work in AI engineering full-time, but most of what I do there stays behind closed doors. This blog is a separate, personal lab notebook: a place to pick up a new model or technique, actually run it, and write down what happened — including the parts that didn\u0026rsquo;t work.\nThe goal isn\u0026rsquo;t polished tutorials. It\u0026rsquo;s closer to a running log:\nwhat I tried what broke, and why what I\u0026rsquo;d do differently next time How this is built Static site with Hugo and the PaperMod theme, deployed to GitHub Pages via GitHub Actions on every push to main. Each post is a self-contained folder under content/posts/, so code output and images from an experiment can be dropped in alongside the write-up.\nA code block, to confirm syntax highlighting works:\nfrom transformers import pipeline generator = pipeline(\u0026#34;text-generation\u0026#34;, model=\u0026#34;gpt2\u0026#34;) print(generator(\u0026#34;The best way to learn a new model is\u0026#34;, max_new_tokens=20)) And inline math, to confirm KaTeX renders: the attention weights in a transformer are computed as\n$$ \\text{Attention}(Q, K, V) = \\text{softmax}\\left(\\frac{QK^\\top}{\\sqrt{d_k}}\\right)V $$\nIf both of those rendered correctly, the scaffolding works. Real experiment write-ups start next.\n","permalink":"https://4oh4.github.io/posts/2026-09-20-starting-this-blog/","summary":"Why I\u0026rsquo;m keeping a public lab notebook, and how it\u0026rsquo;s built.","title":"Starting This Blog"}]