## Computation - Batch ones are easy: just use dbt. - ![[60e69a53 0b87 4b61 81e4 81c512721281]] (https://www.tecton.ai/blog/why-real-time-data-pipelines-are-hard/) - Interesting interplay between [[DBT|Dbt]] and ML for [[scalable-computing-of-features|Scalable Computing Of Features]]: [dbt + Machine Learning: What makes a great baton pass? | dbt Developer Blog](https://docs.getdbt.com/blog/maching-learning-dbt-baton-pass) - Why not with https://www.terality.com/? It may be the easier way out! ## Notes - If we plan on using BigQuery, we need to think of the costs that will have, compared to a DB that does not charge per query or files. - The popular arch reference from [[gcp|Gcp]]: https://www.datacouncil.ai/talks/building-a-feature-platform-to-scale-machine-learning -  ## Refs - [Do You Really Need a Feature Store? | by Lak Lakshmanan | Feb, 2022 | Towards Data Science](https://towardsdatascience.com/do-you-really-need-a-feature-store-e59e3cc666d3) (Do we?)