Writing on AI engineering.

Field notes on building LLM and voice systems — latency, cost, and the unglamorous engineering that makes AI actually ship.

Design Systems

The Master UI/UX Playbook

A single reference for building interfaces that look designed, not defaulted — a portable, token-driven system distilled from auditing several production frontends.

16 min read

AI Systems

Build the system, not the prompt: how I approach an LLM project

The steps I actually follow when turning a vague requirement into a working LLM or voice system — and why the order matters.

5 min read

LLM Engineering

max_tokens isn't free: how a token budget became a 10-minute timeout

The parameter everyone sets once and never thinks about again. Here's the day it cost me an afternoon.

4 min read

LLM Engineering

Stop guessing your LLM bill: per-call token-and-cost accounting

If you can't see what each call costs, you can't make it cheaper. A small wrapper fixes that.

3 min read

Voice AI

Cutting latency in a real-time voice bot: VAD, speculative generation, and a 300 ms rescue

A real-time voice agent is three slow things in a trench coat. Here's where the time goes — and how to claw it back.