Blog
Notes from the build.
Why my homepage is an AI agent, not a contact form
The thinking behind leading with a working agent, what it does (chat, voice, memory), what it deliberately refuses, and why the demo is the portfolio.
23 July 2026 · 3 min read
Most AI bills are an expensive model doing cheap work
The gap between a frontier model and a fast small one is routinely 30–60× per token, and most of what runs through the big one doesn't need it. How I find the difference.
23 July 2026 · 2 min read
You might not need a vector database
For a knowledge base of hundreds of documents, keyword search over well-kept files often beats the embedding stack. Cheaper, debuggable, and every answer traceable to a file.
23 July 2026 · 2 min read