0

I'm working on a project to create an AI Quotation tool using the ChatGPT API trained on my data. The data I have spans the last 10 years and is mostly in Excel and PDF format, but it's not consistently structured. Each entry includes customer name, task required, and the price quoted, along with the list price for all tasks we offer.

However, here's the challenge: most companies have specific requirements for the same tasks, resulting in different prices for different companies. The AI needs to identify the company details, match it with the correct task price, and adjust for inflation if necessary. For example, if a company hasn't had a particular task done in 4 years and we've had 3 price increases during that time, the AI should calculate the correct amount for that company considering inflation.

I've done some research, and it seems that results might be inconsistent due to the varying data formats and complexities. I'm curious if anyone has experience with a similar project or has any ideas on whether this is achievable? Any insights, suggestions, or advice on how to approach this would be greatly appreciated! Thank you in advance for your help!

1
  • 1
    This really doesn't sound like the sort of thing an LLM should be used for.
    – Mark
    Aug 4, 2023 at 7:35

1 Answer 1

3

For example, if a company hasn't had a particular task done in 4 years and we've had 3 price increases during that time, the AI should calculate the correct amount for that company considering inflation.

I'd claim that, especially when reliability is required, using LLMs to perform this task would be a bad fit. Specifically:

  • Unlike calculators, LLMs are fuzzy and don't perform well at precise calculation - particularly with decimals and large numbers

  • LLMs' logical reasoning ability is limited. It probably won't be able to reliably recognise when price increases need to be applied

You could augment the LLM (such as giving it the ability to invoke a calculator API, or feeding in as a prompt the next step it should take - "apply inflation of 3.5%") but at that point it's unclear what value the LLM is adding.

It sounds as though you have a procedure that could in theory be defined in code, and the primary issue is inconsistently structured data. Rather than trying to get an LLM to follow the procedure, I'd recommend parsing the unstructured data into a machine-readable format once and then using more traditional methods to calculate the quote value. This parsing could involve still LLMs, such as OpenAI's example here, with some sanity checks.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.