The original title is "Real-Time vs Batch Indexing: Stop Choosing, Start Combining"

Short summary

The article argues that production search systems need both real-time and batch indexing running simultaneously, not one or the other. Real-time handles the freshness window where stale data causes measurable harm; batch handles historical backfills, embedding refreshes, and schema migrations. The hybrid architecture routes events to both layers, with queries fanning out and merging results, and the freshness boundary is a product decision, not an engineering one.

• •Real-time and batch indexing are complementary, not mutually exclusive
• •Hybrid architecture uses streaming for recent data and batch for historical data
• •Freshness boundary should be a product decision based on user-visible staleness tolerance

Generated with AI, which can make mistakes.