I have a question about AI in KNIME
Is it possible to create an Agent and “tools” that perform a “Replace” on fields within KNIME?
Example:
Imagine I have a table with 10 columns of value, and a lot of rows.
Imagine I need to validate certain rules across some of these columns.
Using KNIME nodes:
If I wanted to validate these columns, I would use the Column Expressions node (which I prefer) and write several IF statements to identify the issues, selecting the “Replace Column” option for each one.
Using AI:
My idea would be to use an LLM from GPT4All, and I would create an agent and “tools” to handle this task.
I am thinking of setting rules for the Agent to determine which tool to use for each type of error.
Obviously, I would have to define all the types of adjustments that need to be identified and corrected.
But my main question is:
After the LLM performs the analysis, is it capable of outputting a “Replace” for each column in the table?
What I commonly see is the LLM producing a summarized document, report analysis, or insights.
However, I want the agent to handle the entire “Replace” process or even create new columns for each iteration of every row.
I am not sure if I was clear in my explanation.
if each column+row is looked at alone, you can simply feed that value as a single call + rule (doesnt make sense but would work).
you can also paste whole rows or columns and retrieve whole blocks, still resulting in many calls.
you can also hand over the full table and return the full table.
depending on your rules and replacements, you can also generate the code and throw it into a column expression or expression node to rework the columns in one step.
if its non-overlapping replacement, string replacer dictionary will work, too
if its just “cleanup” steps, String cleaner will also rework multiple at once.
for numbers, you can also force ranges using the domain nodes.
lastly, you can create and apply PMML models
Sorry, but I don’t understand what you said.
In my case, it’s not about performing ETL.
It involves value movements between fields that are incorrect.
It is a complete analysis of the fields to identify errors.
It’s hard to explain.
But the if-else rules are complex.
I have approximately 300 rules, as each one depends on what is contained in the row.
I’m looking to replace these 300 rules with AI.
consider being less vague or contradicting when you describe your issue.
Imagine I need to validate certain rules across some of these columns.
here you say you want to validate rules
If I wanted to validate these columns,
in the next sentence you want to validate the columns
you talk about
“Replace”
but never bothering explaining what you mean by “Replace”.
but as I have written before, you can pipe your whole table to a LLM, give it your rules and have it spit out a reworked table. If that reworked table actually aligns with your rules is a separate question.
Hi @Richter
That’s it.
That was exactly what I had in mind.
But based on your explanation, it might be complex, and that is precisely what I wanted to know: the level of complexity involved in putting this into practice.