Fetching no, or random wrong data

Created on 22 September 2024, 2 months ago

Problem/Motivation

When enabling AI automator on an address field that uses as a base field a plain text field where I pasted a full set of address information, the result of the ai automation process only fetches:

  • Organization name
  • Street address
  • Postal code
  • Country

It skips suburb, city, and province values.

Furhter down the chain, it also fails to provide any other field data, even if I provide the missing address fields manually, except the "Rating amount" field. But that rating amount is different each time and never the correct amount of ratings from the business. I have tested this with the multi-field where I map all possible fields, and on an individual phone field.

Steps to reproduce

Proposed resolution

Remaining tasks

User interface changes

API changes

Data model changes

🐛 Bug report
Status

Active

Version

1.0

Component

Code

Created by

🇹🇭Thailand AlfTheCat

Live updates comments and jobs are added and updated live.
Sign in to follow issues

Comments & Activities

  • Issue created by @AlfTheCat
  • 🇩🇪Germany marcus_johansson

    Could you give an example of an address that has suburb and province values? They have never worked, so its rather a feature request, but easy to add. City I just tried and it worked, but if you give me an example of a none-working city it would be great.

    Also in my test it fetches everything if you assign it correctly. See screenshot below.

    So all in all, I need more information on what goes wrong and why it goes wrong to be able to replicate it? Could you perhaps write down the steps you took to cause the failure and the places you searched for (or similar places if its a business secret)

  • 🇹🇭Thailand AlfTheCat

    Hi Marcus,

    Thanks for your attention to this.

    In my setup I first interpolate the address from a plain text field using the "Google Places: LLM Text Addressfinder"
    I realized that perhaps I was using the wrong AI model so I changed the advanced settings in the field config to GPT4-o and that yielded the correct address for my test search on "Eiffel tower Paris".

    The multi-field however, still does not fetch any data. I noticed there is no "Advanced" fieldset there, so I can't change the model in the same way as the Address field. I changed the default in the AI module setting for "chat" to GPT4-o which did not change the results. The "chat with complex json" default was already set to GPT4-o.

    There are no errors in the log, I am running the interpolator via batch and it does run, briefly, but doesn't yield any data on the address of the Eiffel tower using the attached address data.

  • 🇩🇪Germany marcus_johansson

    Which version of the AI module, Drupal and PHP do you run?

  • 🇹🇭Thailand AlfTheCat

    I'm on PHP 8.2, Drupal 10.3.5 and I should be on the latest or near-latest dev release of the ai module, I updated it a few days ago.

  • 🇩🇪Germany marcus_johansson

    Ah, sorry, understood now - did you setup the mapping of the data to different fields? See the screenshot:

    If its setup like this and it still doesn't work, you can expose a checkbox that reruns everything, by going to Manage Form Display, place the Google Places Multifield in the form and then click the cogwheen and uncheck "Hide the field completely.". That way you can rerun it over and over.

  • 🇹🇭Thailand AlfTheCat

    Hi Marcus, thanks for all the attention to this. It turned out that the Google Places API key was causing the problem. I don't know why, because the key is active, but I found an old one I created 3 years before, and using that key did the trick. It's working beautifully now!

    Maybe it's useful, if possible, for the module to write more watchdog information in case of failures to make it easier to debug. Because it was working in part (filling out the address field worked), I didn't suspect the API key as the source of the problem.

    Fantastic module, very happy to have it working!

  • Automatically closed - issue fixed for 2 weeks with no activity.

Production build 0.71.5 2024