Tool
Saudi tender tagger
Paste an Arabic tender title from Etimad. The model I trained reads it and guesses what kind of work it is.
About this model
I trained it on more than 7,000 public Etimad tenders. On tenders it never saw, it puts the right activity first 61% of the time and inside its top three 80% of the time, out of 54 activity labels. It gets the procurement type right 88% of the time.
Use it to screen a long list so a person looks at the likely ones first. Do not use it to decide what to bid on. It will be wrong on about four titles in ten, and the confidence bars are the model’s own, not a promise.