Ideas for 0.3.0

[blythed]

• Time-series analysis
• ...

[blythed]

Time-series analysis will require thought around how to update when new data comes in. (No longer row-by-row due to dependency on previous data-points.)

[fnikolai]

Decide the AI and system metrics that makes sense to expose to Prometheu.

[fnikolai]

Integration with ElasticSearch.

[jieguangzhou]

• Unified interface for MongoDB and SQL.
• Automatic datatype recognition eliminates the need for manual data configuration in the vast majority of cases.
• Support model plugin package to make the main repository simpler (for example: pip install superduperdb-ollama, then we can use ollama; pip install superduperdb-pgvector, then we can use pgvector)
• Support the predict function of all models to run on the computing service.

[blythed]

Hmm, we had something like this, but then moved away from it. Can consider re-introducing (plugins).

[blythed]

• Choose to serve any model on ray serve instead of in-process (agree with @jieguangzhou)
• Make db.apply more sophisticated, allowing for diffs to be taken into account
• General disaster recovery, locks (e.g. during training) and failsafes
• Broaden support for async "under-the-hood" for API calls
• Generally make the developer contract easier
• Some integration model for string-only query mark-up "SELECT * FROM docs WHERE docs.txt LIKE 'this is a test'"