DaedalusPixAI Feature Platform

A YAML-driven feature catalog, a DuckDB/Polars rolling-aggregation engine, an operator-pipeline training path, and a lean platform layer — all Arrow-native.

Get Started

What is Daedalus?

View on GitHub

📚

Declarative Feature Catalog

Define feature views in YAML; FeatureCatalog parses DuckDB-like dtypes into Arrow and resolves entities, lineage, and column-level service contracts.

⚙️

Dual Compute Engines

Point-in-time rolling aggregations on DuckDB or Polars — bit-for-bit equivalent — behind one engine factory.

🧬

Operator-Pipeline Training

One canonical skeleton → enrich DAG of engine-tagged operators, run through the executor via `daeda pipeline train`.

🏹

Arrow-Native, Streaming

Lazy, streaming, vectorized batch (SIMD) compute end to end — Ray Data over Arrow, no pandas in the hot path.

🗄️

Pluggable Data Sources

Config-only backends — Parquet, DuckLake (1.0 ATTACH), Postgres, and optional Snowflake (key-pair JWT, query pushdown).

🛰️

Platform Surfaces

Dagster orchestration, a FastAPI JSON-RPC control plane, an MCP server, an agent CLI, and an optional MLflow projection — all over one in-process API.