[rules-users] Performance with 'eval'

Hi. I need to implement some sort of 'adviser' or 'recommender' system, and I'm considering JBoss Rules to do the task. It would be an adviser for choosing which car to buy, for example. Imagine: 'I want a car which is all-wheel-drive, and I would like it in green, but I don't mind if it's in another color'. The user can choose how important is each feature for him, or if it's mandatory (in the previous example, the 'all-wheel-drive' requirement is mandatory, while the color it's not). So, I would have all the car catalog in the working memory permanently (class Car), and I would assert a buyer (class Buyer), and I would like to get the cars that best match the user requirements and wishes. I had outlined something like this (pseudocode): ################ rule "CarRecommendation" when buyer : Buyer() car : Car (feature1=mandatoryRequirement1, feature2=mandatoryRequirement2, ...) eval (carUserCompatibilityLevel(car, buyer) > COMPATIBILITY_THRESHOLD) then buyer.getRecommendedCars().add(car); end function int carUserCompatibilityLevel(Car car, Buyer buyer) { return [compatibility level based on car suitableness to user preferences]; } ################ My main question is if I'm forced to use an 'eval' element in order to get what I want. I'm afraid I need to use it, and I'm concerned about the performance since the WorkingMemory will hold up to 10 million 'cars' (well, they aren't really cars, the car-buyer thing it's just an example for the sake of simplicity), and the docs clearly state that evals are not nearly as efficient as field constraints (memory usage is not a concern in this moment). Since the 'car' catalog is so large, I wish I could get the cars with the highest compatibility levels first and stop when I reach, for example, 50 recommended cars (I would add something like 'if buyer.getRecommendedCars().size() > CAR_LIMIT retract(buyer)' to the consequence), and so I would't have the need for a COMPATIBILITY_THRESHOLD. I don't know if that's even possible. I'm new to this kind of programming and any tip would be appreciated, even if it involves to totally change the approach (i. e., not using a rule engine for the task). Just any clue that points me in the right direction. Thanks in advance