- building outputs for price fairness (based on publicly available labor data)
- scope match (is vendor over/under scoping user's intent)
- risk (vendor risk, timeline, price variability, etc.)
- value (some combination of price, service, longevity, etc.)
I don't get much hallucinations in my testing, but overall it's pretty complex pipeline since it is broken down into so many steps.