Hi Everyone,
I am currently working on an internal application for editing and analyzing forecast data built and hosted via Databricks Apps. I would appreciate a sanity check from anyone who has taken AppKit into production as I am planning on utilizing the official AppKit TypeScript SDK to leverage pre-built plugins, alongside agent orchestration using LangChain Python SDK behind a Model Serving endpoint.
The users would ask an LLM-driven conversational agent questions about forecast data, based on which the agent would be able to query the relevant data and get back tables and charts. The trickier part is that application should allow users to edit and modify the returned data either by natural-language prompts or directly in an interactive table, and then those overrides should be written back to the underlying tables.
Are there any established AppKit patterns or plugins for transactional write-back and optimistic concurrency control and audit history, or more bespoke solutions are needed. My current thinking is to use Lakebase via its plugin and an append-only history table. Perhaps a Python-based architecture (e.g. Dash) with hand-build flows would be more suitable?
Thanks in advance for any architecture examples, lessons learned, or documentation pointers.
