What problems is TrueFoundry solving and how is that benefiting you?
It allows our ML engineer (which are rarely knowledgeable about those kinds of things) to deploy and monitor their applications themselves. This avoids the anti-pattern of ML teams throwing their models to a separate Ops team. But it still allows fine-grain control, when an Ops specialist needs to jump in.
It made it easy for us to implement Continuous Delivery (we auto-deploy to our dev environment every time a push on origin/main changes a transitive dependency of a service).
It allowed us to reduce our computing costs in several way:
- It made it easier to use spot instances for our dev and staging environment, reducing the costs of those environments by 80%.
- It made it easy to turn off our most expensive services during nights and weekend (we only need them during office hours).
- It reduced resource waste, by making it easier to run multiple small services on the same machine in an isolated way.
It still feels we are under-using True Foundry, since we are mostly deploying "services" and haven't deployed jobs or notebooks yet. Review collected by and hosted on G2.com.