llermaly
commented
9 months ago Ideally we want something like this:
Input:
Given this summary: the car shows damage in the X zone, Y Zone" This replacements needed ["left door", "right door", "front mirror"]
I found the following product (or products?) in the store:
LEFT DOOR SKU 1111 RIGHT DOOR SKU 1234 BACK DOOR SKU 84213 TIRES SKU 842424 BREAKS SKU 88984 .......
Which ones are required to cover what the summary and what the replacements needed are asking?
Output:
"left door", LEFT DOOR SKU 1111 "right door": RIGHT DOOR SKU 1234 "front mirror": NOT FOUND
We can then go to the database and search back to get the details.
Once we send the picture and get the details from the damaged parts we go to the database to get the potential parts needed.
By definition we may get more parts than we need. So, we need to exclude the parts we don't need using a refinement process.
We can use the evaluation for this, sending the part (or parts) + the damaged parts the LLM sent on the first place.
I prefer to send the parts separately so we can also send some descriptions to help the model decide.