Share
ReportHi everyone,
Imagine you have an investment application form and you are building an agent that helps users formulate answers for this form.
The challenge is that each answer has a strict character limit.
The agent should:
- ask the user for input,
- compare the input against available best practices,
- generate a well-formulated answer,
- make sure the answer does not exceed, let’s say, 2,000 characters including spaces,
- then continue question by question.
- ask the user for input,
- compare the input against available best practices,
- generate a well-formulated answer,
- make sure the answer does not exceed, let’s say, 2,000 characters including spaces,
- then continue question by question.
The problem is that we know LLMs cannot reliably count characters. They usually estimate it.
How would you design such a solution from a character-counting and answer-adjustment perspective?
I tried one approach with Generative Answers: the agent generates an answer block, measures the length, and if the length is not acceptable, the process repeats a few times in a loop.
But I am not fully happy with the answer quality when using Generative Answers.
A custom prompt as a tool also tends to hallucinate too much.
If I use a child agent, the answer quality is better, but then it is not very clear to me how to store the AI-generated answer in a variable and measure its length.
Any ideas? ^^
I have the same question (0)
All responses (
Answers (
2,665
