Insights on cloud architecture, distributed systems, AI, and modern software engineering practices.
Elasticsearch supports five distinct search layers, from BM25 to Agent Builder. Understanding when to use each layer is the difference between a search that works and one that scales.
Better Binary Quantization cuts dense vector memory by 96% and DiskBBQ eliminates the RAM ceiling entirely. Here is how both features evolved across Elasticsearch 9.x and what changed in 9.5.
Elasticsearch 9.5 ships the semantic field type as GA, auto-calibrated DiskBBQ, batched query execution, ES|QL Data Federation, and a major ECK operator update. Here is what changed and what it means operationally.
Air-gapped semantic search with Elasticsearch and Jina AI — no outbound calls, full data residency. Sovereign cloud IaaS hits $80B in 2026 (Gartner). One command deploys the stack.
Cut Elasticsearch storage by up to 95% using mapping discipline, ILM tiering, LogsDB, and pattern_text — a practical guide for platform engineers.
Elasticsearch 9.1–9.3 cut vector memory 95%, landed DiskBBQ sub-20ms search, made ES|QL production-ready, and acquired Jina AI. Here's what changed operationally.