Blog
Notes on AI infrastructure, LLM engineering, and what I learn along the way.
MECE
A messy list is usually two questions pretending to be one. MECE is just a name for noticing that.
Thinking·3 min read
Radix Cache
Many requests share the same prefix. A radix cache keeps that work around so the next request does not pay for it again.
LLM Engineering·8 min readThe model is just the beginning
Any team can wire up an LLM API now. The interesting engineering starts after the demo works.
Platform Engineering·2 min read