ls -la posts/ | head -n 5

• 2026-04-08 [code-gen] From Jira to Pull Request: Building an AI Bug-Fix Agent llm ai-engineering code-gen
  In Part 1 I built a Copilot clone for inline code completion. In Part 2 I built an NL2Code agent that generates full scripts from English descriptions. Both of those...
• 2026-03-09 [code-gen] NL2Code with Agents: From English to Production Python llm ai-engineering code-gen agents
  In Part 1 I built a Copilot clone that does inline code completion. Fine-tuned it on our codebase, got it generating completions that actually follow our patterns. That was the...
• 2026-03-04 [code-gen] Build Your Own Copilot in Pure Python llm ai-engineering code-gen
  In the skills post I mentioned we’d been building agent skills at work, mostly for marketing ML workflows. One of those workflows is NL2Code: a user describes what they want...
• 2026-02-28 [transformers] Is Attention Really All You Need? deep-learning llm
  A couple weeks ago I wrote a walkthrough of Karpathy’s microgpt.py, tracing every line of a GPT implementation in pure Python. While putting that post together, I kept wanting to...
• 2026-02-16 [transformers] GPT in 200 Lines of Pure Python: Every Line Explained deep-learning llm
  I was scrolling through X late on a Friday night (as one does) when I spotted a post from Andrej Karpathy linking to a new gist. The title: “The most...
• 2026-02-05 [agents] The Skill is the Unit of Reuse llm ai-engineering agents
  In the last post I talked about A2A, MCP, and where multi-agent systems break down. I ended with a teaser about the “single agent with skills” pattern being the sweet...
• 2026-01-28 [agents] The Agentic Protocol Zoo llm ai-engineering agents
  I’ve been building agents at work for the past few months. Not the “wrap an LLM in a while loop and call it an agent” kind, but actual multi-step systems...
• 2026-01-12 Here Be Dragons meta ai-engineering
  I spent the holidays in France with my family: the usual routine of too much food, bad Wi-Fi, and long walks through the countryside. But this time was different. Normally...
• 2025-12-28 [fine-tuning] LoRA From the Ground Up: The Math, the Matrices, and the Merge llm fine-tuning
  In my last post, I walked through SFT, DPO, and RLHF for fine-tuning LLMs. Throughout that entire post, LoRA kept showing up in every code example, every training config, every...
• 2025-12-22 [fine-tuning] SFT, DPO, RLHF: Picking the Right Fine-Tuning Strategy for Your LLM llm fine-tuning
  I’ve been spending a lot of time with Llama models at work lately, mostly prototyping, seeing where the edges are, getting a feel for what a mid-size open model can...
• 2025-12-15 Welcome to the AI Terminal meta announcement
  This blog is a space for going deep on AI/ML: the engineering, the research, and everything in between. If you’re here, you probably care about how things actually work under...