We built an MLOps platform to offer real-time collaborative Jupyter notebooks and you could train, track, deploy, and monitor your ML models...
Here's something I posted some time ago: https://news.ycombinator.com/item?id=29397405
>Some of our earliest design decisions were:
>- Everything one could do with point and click must be possible with an API call. A developer should build a clone using the API. Developers should be able to build products on top of this or use it as a subsystem in their systems.
>- Build on top of common abstractions and units of 'thought'. Docker, S3, HTTP, JSON, etc.*
>- The platform must be able to die without users left scrambling to exfiltrate their work (based on my personal motto of being able to die without the team being impacted beyond nostalgia).
I laid these out and enforced them. That's why instead of hosting user data, I insisted we integrate with their S3 buckets and their Kubernetes clusters. Users could even use the data on S3 like a file-system from their notebooks (read_csv instead of boto3). Not to mention that it's easier to sell into the enterprise if you do that (CISOs, CTOs, CFOs, etc).
Have you tried this: https://x.com/PaperspaceOps