Performance Articles
Performance engineering for analytics: Query optimization, columnar storage, and vectorized execution. Real benchmarks and production latency wins.
Know What a Query Reads Before It Runs
Row estimates only hint at what an agent will read. Bytes pin it down. Our explain returns them from catalog metadata, before the engine touches a file.
Grep your lakehouse
Agents fail when they cannot retrieve the right data slice before writing SQL—not because they cannot generate queries.
AI's Event Backbone
AI-native products generate a new kind of infrastructure problem. Here's how to build the event backbone for your AI system.
One Billion Rows
At 1 billion rows, every shortcut comes back to collect interest. Here's how we achieved sub-second queries with near-realtime ingestion.
Pruning Top-N Queries
A deep dive into DuckLake PR #668 and how Top-N dynamic filter pruning turns ORDER BY + LIMIT from full scans into metadata-driven execution.
Lessons from Search
Real-time analytics faces the small-file problem search engines solved. DuckLake's tiered compaction brings those merge strategies to streaming analytics.
Stop Batching Analytics
Why we're forcing analytics through complex batch pipelines when append-only data should work like logs. The warehouse constraint that stopped making sense.
Speed Shapes Understanding
Speed isn't just a luxury: it's the difference between insight and inertia. We've been deep in TPC-H benchmarks, tuning our analytical engine for AI agents.
NetQuack 4000x Faster
We rewrote NetQuack DuckDB extension, replacing regex with character parsing. Result: 4000x faster—37 seconds down to 0.012 seconds.
Rethinking the Lakehouse
Breaking down our storage and query architecture: why we're leaning into Apache Iceberg and why DuckDB is emerging as our real-time query engine of choice.