Teaching HeavyIQ
Sometimes HeavyIQ picks the wrong column for a certain value, maybe because the column name remember a concepts that eventually resides in a different column, or it is too much generic and unrelated.
While this is an expected behavior, I was wondering if there is a way to "teach" HeavyIQ that when I ask for a metric, it/she/he ;) has to use the column, or expression, that I pointed out.
Something like a "tag" or similar, for example, the tag "gross_incoming" for a "totgin" named column.
In this case the column name would not help HeavyIQ in picking the right column, unless I glue the tag "gross_income" to the "totgin" column. This is the idea: is this possible?
-
Official comment
Hi Gianfranco,
This is a great question. 8.0 supports tagging metadata to tables and columns, and that metadata is injected into the HeavyIQ prompts to help give the model more context on the data it is looking at.
For example, you can write something like
`COMMENT ON COLUMN weather.temp IS 'Temperature in Celsius';`
And it should get picked up in the prompt. Currently you need to do this in the SQL editor, but the next major release of Heavy.ai will allow you to do it via a GUI in the Data Manager in Immerse.
We're also working on a full RAG capability that will allow arbitrary upload of "facts" and documents, that will get selectively injected into the prompt depending on the user question, which will make it easier to add facts that are not column or even table specific (like KPI definitions).Comment actions
Please sign in to leave a comment.
Comments
3 comments