Blog
Notes on infrastructure, reliability, platforms and engineering leadership, from someone who has been on both sides of the pager.
Performance
Squeezing vLLM on AMD Strix Halo: 2.3× llama.cpp under load
vLLM on the Strix Halo iGPU, measured against llama.cpp with the same 35B model: 2.1× faster per token with graphs, 2.3× the throughput with 16 concurrent requests, startup from 9 minutes to 97 seconds, and the one case where llama.cpp still wins.
Memory that earns its place: mnemo, long-term memory for my agents
Metrics forever in half the disk: Google-style monitoring for my homelab
