Grouper Training - AI - Lesson: Making an agent

Getting started

Connect to your VM

Maturity level 4 - custom GPT

1. In ChatGPT, click on explore under GPTs on left

2. Click Create

3. Make sure you are in the Create tab on left

4. Type in Create instructions (on left side)
   
   

   Your purpose is to process Grouper subject IDs and generate a GSH script for the user to download. Ask the user to upload a CSV of Grouper subject IDs. This upload file is required. If there is a header in CSV file, ignore it. Use a script to process the the CSV file to look at each subject ID and it is numeric, if it is odd, output a Grouper GSH line to add the subject ID to group: test:testGroup. If it is even, the GSH script line should add the subject ID to test:testGroup2. Use method chaining in the GSH script. Remind the user to start their GSH terminal by running "./gte-gsh" in their Grouper Training Environment terminal.

   
   
   

5. It will ask about name, type this
   
   

   Name: Grouper test even odd

   
   
   

6. Type "finish" and accept prompts

7. Click "Configure" tab

8. Notice how the instructions looks different from the above text.  Let AI configure AI.

9. Keep one Conversation Starter only (if it generates more, delete the extra ones at random)

10. Under capabilities select only "Code interpreter and data analysis". Uncheck the other boxes

11. Click "Create" in upper right, save (don't worry about who it is shared with)

12. Use the GPT, click the prompt button

14. Edit the GPT (GPTs > Explore > My GPTS) and iterate, maybe tell it (in configure → instructions, or in create tab).  Make sure prompts are same if using Create tab 
    
    

    do not to fail if the membership exists, e.g. grouperSession = GrouperSession.startRootSession(); testGroup = GroupFinder.findByName(grouperSession, "test:testGroup"); testGroup.addMember(SubjectFinder.findById("800001115"), false);

    
    
    

15. Click "Update" in upper right and try it again

16. Was something wrong (e.g. how to run the script)?  Edit it and explore

Maturity level 5 - RAG

1. To find a course group, you need to know course codes and descriptions, maybe you want a better search

2. Make a new GPT, called "Grouper object search"

3. Instructions
   
   

   Use the gte_groups_folders.json knowledge file to search for groups and folders as the user requests. Ask the user what they would like to find, and the JSON data for groups and folders has the answer. If the result is a group, you can give the user the group extension (last part after the last colon in the name) in a link to this URL: https://localhost:8443/grouper/grouperUi/app/UiV2Main.index?operation=UiV2Group.viewGroup&groupName=THENAME or for folders (aka stems): https://localhost:8443/grouper/grouperUi/app/UiV2Main.index?operation=UiV2Stem.viewStem&stemName=THENAME The gte_groups_folders.json has a JSON array with three attributes: type (group or folder), name, and description. Use all of those to search as the user requests.

   
   
   

4. The only capability is "code interpreter and data analysis"

5.  to the GPT

   1. That file was exported for you from a SQL query
      

      
      
      

6. Use GPT 5.1 model

7. Save the GPT

8. Try this prompt into the GPT:
   
   

   What religion course groups of professors are there?

   
   
   

9. If it doesn't find them, ask to look again

10. Lets tune up the dataset.  We will have AI assigned some keywords for the groups and folders in the JSON.  If there is a menial task, we want AI to do it.

11. Click on the upper left icon to get to ChatGPT not in a custom GPT

12. Use some prompts and upload the  knowledge file.  It takes some back and forth, so be creative, here is an approach that works
    
    

    Adjust this json file. Take the group name and assume some keywords. For example if the group name is: basis:sis:course:rel:rel301:student then the keywords could be: religion, theology, church, faith. Add a JSON attribute called "keywords", and put the comma separated keywords in the value. Then give me the new JSON file to download.
    the keywords attribute is mostly blank, please put comma separated values in there based on the name of the group or folder. each one should have at least 1 and max 5 keywords
    full nlp extraction. the keywords are different for each array item. This is an example of the JSON file { "grouper_objects": [ { "type": "group", "name": "etc:privilege:stemViewPrecompute", "description": "If you are having performance issues with UI users and stem privileges, put users in group who should be precomputed in the stem view full sync daemon", "keywords": "Performance, UI, folder, privileges" }, { "type": "group", "name": "etc:sysadmingroup", "description": "system administrators with all privileges" },
    do it again and only have keywords for the groups and stems for courses in the student information system. take the three or four letter codes and assume what subject it is and make keywords based on the that. e.g. "bus" is assumed to be business. "soc" is assumed to be sociology, etc.

    
    
    
    Might look something like
    

    
    
    

13. Edit the GPT and:

    1. Click Grouper Object Search on left under GPTs

    2. Click the drop down in upper left and click Edit GPT

    3. On the configure tab, delete the existing knowledge file

    4. Upload the new knowledge file

    5. Change the instructions to reference new file name (whatever ChatGPT named it for you to download), it could be: gte_groups_folders_sis_course_keywords.json

    6. Explain the keywords attribute

14. Save the GPT (click Update button), and try the prompt again:
    
    

    What religion course groups of professors are there?

    
    Maybe try this prompt
    
    

    groups of instructors for classes of religious topics

    
    
    Should find 6-8 groups, the more the better!

Sample resulting instruction set