Search for a command to run...
Series
The field study series. Flake-sleuth (21 repos with CI logs), incident-commander (15 outages), rag-oncall-chunking (3 strategies benchmarked), tierforge (cost tracking across 4 models). The honest post-mortem format you already nail: setup → what happened → what broke → what I'd change.
TL;DR — 14-node LangGraph pipeline. 521 tests. 88.3% field test pass rate across 15 real incidents. $0 to run on a local LLM. The biggest surprise: my test design was wrong, not the tool. The 2 AM P
TL;DR — I built a RAG assistant for on-call engineers that indexes 55 runbooks. Header-split chunking hit MRR 0.9524 vs 0.8893 for fixed-size. The embedding model, LLM, and vector DB never changed. Ch
TL;DR — AI agents get stuck in loops. A $0.001 call that loops 10× costs more than a $0.05 call that solves it once. I built ai-loopguard, a drop-in circuit breaker that detects stuck patterns and esc
