How Seltz cut data costs 40% with Amazon S3 Tables

This article describes how Seltz, an AI search engine company, achieved 40% cost savings by adopting Amazon S3 Tables for its knowledge pipeline that processes billions of documents through enrichment stages.

• S3 Tables provide fully managed Apache Iceberg tables with automatic compaction, snapshot management, and file cleanup
• Seltz's pipeline uses multiple enrichment stages (text extraction, quality scoring, deduplication, vectorization) that write millions of small files requiring compaction
• Managed lifecycle eliminates operational overhead of building custom maintenance pipelines with Spark or AWS Glue
• Atomic commits and time-ordered IDs enable incremental processing where only affected downstream stages reprocess data
• Amazon Athena and DuckDB provide ad-hoc SQL access for debugging and monitoring without additional ETL
• Step-by-step walkthrough covers creating table buckets, writing data with PyIceberg, querying with Athena, and reading programmatically

S3 Tables eliminate engineering complexity for managing Iceberg tables at scale while preserving open format flexibility and enabling observable pipelines.