I’m currently building a job-recommendation bot for people who are unsure which career they should get into.
I basically have a list of 100 Professions (JSON format with ‘name’ and ‘id’), and intend to have the assistant ask questions like ‘Which field are you most interested in? Technology - Retail - Construction’, and then have the assistant return the best-fitting professions.
The jobs are uploaded in the file ‘available-jobs.json’ and I use the function ‘job_proposals_issue’ to eventually out put the desired ids.
However, even though I explicitly tell the AI to only include ids, which are available in the desired file, it frequently returns ids, which are not.
I also tried adding the JSON object of the available jobs into the function description itself, however, it still missbehaves frequently.
Is there something I’m missing?
Is there a better way to create an agent which creates well-fitting functions around a specific set of possible values and outputs them accordingly?
This is what the function looks like:
{
"name": "job_proposals_issue",
"description": "Outputs the IDs of the proposed training courses to be displayed to the user.",
"parameters": {
"type": "object",
"properties": {
"message": {
"type": "string",
"description": "Message sent by the chatbot"
},
"jobIds": {
"type": "array",
"items": {
"type": "string",
"description": "ID of a training program to be suggested to the user. This must be a valid ID, i.e. it must come from the available training courses in the 'available-jobs.json' file. If one of the jobIds is not in 'available-jobs.json', this is considered a very serious error. It is better to output fewer ids than incorrect ones."
},
"description": "IDs of the training courses to be suggested to the user. 1-5 of these trainings can be suggested. Make sure that the IDs are valid."
}
},
"required": []
}
}