The main obstacles to a truly open‑source AI stack are fragmented tooling, limited reproducibility, and the scarcity of mature pipelines for edge deployment. The core of our approach overcomes these challenges with a multi‑gateway architecture that decouples the model graph from the execution environment, allowing a single logical model to run on multiple targets while keeping data logic, storage format, and runtime provider clearly separated. Built on AIONIO’s DuckLake + SQLMesh framework, the platform can be reproduced effortlessly on a developer’s laptop and in the cloud—all from one codebase. In the cloud, both the compute jobs and the database are deployed as serverless services, providing automatic scaling without any server‑management overhead. Compared with the Snowflake + dbt combination, this open‑source stack is better suited for heavy data‑transformation and machine‑learning workloads.

The framework equips organisations with a concrete scheme and a curated set of modular tools that can be seamlessly integrated. It enables teams to move quickly from local development to prototyping and then to production‑ready, scalable pipelines. Analytical workloads are handled through clearly defined ingestion, transformation, and layering stages, ensuring robust governance across the entire data lifecycle.