Atlas is a self-hostable analytics engine designed around columnar storage and cookieless identity. It answers arbitrary funnel and retention queries over billions of rows without a data warehouse.
The ingestion path batches events into immutable Parquet segments, while a query planner rewrites funnel definitions into a single vectorized scan. Typical funnel queries over 90 days of data return in under 400ms.
I built the query planner, the segment compaction scheduler, and the dashboard query API.