This project is for mathematicians, programmers and experts in ALGOs. Build me a tool which can value a precious metal portfolio

I am looking to create a bullion trading website. This is a website where buyers and sellers of bars and coins can buy/sell amongst each other to avoid having to pay retail prices that bullion dealers charge. Furthermore if I have a coin collection, the bullion dealer buyback prices are not competitive. This website will bring buyers and sellers together. This valuation project is a side project within the bullion trading website. Many collectors of coins like myself have many many varieties of mints/years/sizes/types etc but struggle to value them in the absence of actual market data. All I have to go on is the price I can buy the coin from a bullion dealer and the price the bullion dealer is willing to buyback the coin. This is a wide spread and the true market price is somewhere inbetween this. Within my website, I want users to be able to build an inventory of their bars and coins and I want to offer a bespoke valuation tool which gives users an approximation of the value of their inventory. The lower bound of the value will depend on the spot price of the metal. The upperbound will depend on a variety of factors. In the simplest version of this tool, if I have 1 coin in my inventory and that exact coin traded on the website on a particular day, the value of my coin will be a function of the traded price (taking into account where the spot price of the underlying metal was at the time). But if I have 1 coin in my inventory and that coin did not trade on my website, then how do I value it? Lets say another coin traded from the same mint but a different year, that information should be the next best thing and used as an input. Lets say no coins from the same mint traded on the day but a coin traded yesterday. That info might be better to use than something else. What I mean is, there will be lots and lots of information about coins trading, there will be information from bullion dealers, there will be historical information, there will be correlation matrices between coins, .....all this information should be used to come up with a strong model to give a fairly good range. As the market data gets more richer and you have more and more price points, the valuation accuracy should increase. I am looking for someone with an idea of how this might work, what the framework should look like, how the rules will work within an algo etc. Of course there will be some coins which are very hard to value. But the majority should use some robust rules to value them and the model should be able to react to new information.