Hi, I recently asked a question regarding the NLSQLTableQueryEngine (https://docs.llamaindex.ai/en/stable/api_reference/query_engine/NL_SQL_table/) and how can I implement the row retrieval part in this notebook (https://docs.llamaindex.ai/en/stable/examples/pipeline/query_pipeline_sql/#2-advanced-capability-2-text-to-sql-with-query-time-row-retrieval-along-with-table-retrieval) but after converting it to workflows instead, and why it is not showing up on Arize Pheonix properly. Anyway, I have made a new class called NLSQLTableRowQueryEngine that inherits from NLSQLTableQueryEngine and a NLSQLRowRetriever that inherits from NLSQLRetriever with small changes to implement the up-mentioned technique in the notebook. I wanted to know what the process is if I wanted to contribute these changes to code base?

Hi everyone,
I had the chance to watch this great video by Jerry (https://www.youtube.com/watch?v=L1o1VPVfbb0) and although Query Pipelines are now depricated and now replaced with Workflows, the same notebook has been updated with Workflows here (https://docs.llamaindex.ai/en/stable/examples/workflow/advanced_text_to_sql/) however, here is the catch:

1. I tried to use Arize Pheonix to trace the execution of the last workflow and what I got was separate record for each step in the Workflow, not nested or linked together.
2. There is a class called "NLSQLTableQueryEngine" which implements the first Workflow in the previous notebook, this one doesn't use Workflows and is tracked very well on Arize Pheonix.

So I wonder if the complete query-time row retrieved implementation can be implemented as a traceable class like "NLSQLTableQueryEngine" or if the Workflow can be traced using Arize Pheonix.

TBH, it's hard to read the source code and understand how everything works (I tried) to answer this question myself.