RAG evaluation: fix retrieval before you blame the model
Most RAG failures are retrieval failures. A practical playbook for retrieval evaluation: golden sets, recall@k, chunking tests and drift monitoring.
Pillar
Scalable AI architecture
The infrastructure underneath AI features: retrieval, vector stores, caching, sharding and latency budgets. Each piece works through the arithmetic and the failure modes before the tooling.
Embedding model migration: re-indexing 10M+ vectors without downtime
A better embedding model is only an upgrade if you can install it. The sizing, dual-write and cutover plan for re-indexing 10M+ vectors with no downtime.
Redis sharding strategy: when a single node stops being the answer
Most Redis sharding goes wrong before the first shard exists. Diagnose the real ceiling, design keys around hash slots, and reshard without a p99 spike.