Blog
Notes from building queue-driven and AI-native backends at scale - distributed systems, Kubernetes, performance, and the failures worth writing down.
12 min read
Vector Search Is a Strategy, Not an Architecture
Why cosine similarity can't answer causal, temporal, or ordered questions - and how Karpathy's LLM Wiki reframes retrieval as write-time compilation, seen from building Cortex's seven-strategy retriever.
RAGVector SearchKnowledge GraphsLLMRetrieval
9 min read
Cutting Kubernetes Cluster RAM by 50%: What Actually Caused the Bloat
How a responsibility split halved a Kubernetes cluster's RAM - why long-running workers pay for imports they never use, and how to find the same bloat in yours.
KubernetesPythonDistributed SystemsPerformanceCelery