This is awesome, and it's pretty difficult. I developed something recently for a specific problem, and it's annoying how easy it is to find yourself writing out an ETL pipeline essentially to enable the LLM when the goal was to enable an ensemble of LLMs to "figure it out." You have to constantly push back. It was essentially grabbing a line from a datatable and querying fairly messy data dumped from PDF files and websites, allowing fuzzy matches with a rank, taking a subset, and then comparing the information between these dissimilar documents to find one to three specific pieces of information. It then updates a datatable for machine learning models. It was a challenge and required a more intelligent LLM (unless you're going to be a software engineer for every single step in detail... then why do Agentic at all?). The solution cost about $9 and was pretty slow, but a human would have taken about six months. Do you think there are lots of people that are basically writing entire codebases of SQL queries, where an LLM becomes an unnecessary step or layer in the SQL query that a human spends writing from beginning to end, which is not what they wanted. The programmer becomes the servant to the LLM.
This is awesome, and it's pretty difficult. I developed something recently for a specific problem, and it's annoying how easy it is to find yourself writing out an ETL pipeline essentially to enable the LLM when the goal was to enable an ensemble of LLMs to "figure it out." You have to constantly push back.
It was essentially grabbing a line from a datatable and querying fairly messy data dumped from PDF files and websites, allowing fuzzy matches with a rank, taking a subset, and then comparing the information between these dissimilar documents to find one to three specific pieces of information. It then updates a datatable for machine learning models.
It was a challenge and required a more intelligent LLM (unless you're going to be a software engineer for every single step in detail... then why do Agentic at all?). The solution cost about $9 and was pretty slow, but a human would have taken about six months.
Do you think there are lots of people that are basically writing entire codebases of SQL queries, where an LLM becomes an unnecessary step or layer in the SQL query that a human spends writing from beginning to end, which is not what they wanted. The programmer becomes the servant to the LLM.