Hi Stacy,

Sorry for the delay.

That's good news the model does work.

I'll guess the service needs to model data by both location ( eg SF, NY or .. ) AND  price ranges for those markets. A 3 BR, 2 BA home in MA would sell very quickly compared to a high end 5 BR, 5 BA home over $1M in most MA markets ( except greater Boston ). If your model allows those inputs, I expect it could be very accurate after training in a market.

Cluster analysis by market ( a set of zip codes ) would show roughly what is in demand on the sales side and what is available on the supply side to do basic matching in a market.

In theory, the only thing that "breaks your model" at any point are black swan events like this virus. I don't know if there would be a "factor" you could apply to the forecasts at that point or if you need a new model during that time period.

I linked a few sites that provide real estate apis you can access data.
Some of those I believe are free.
I don't know if they have the minimum data you need on listing date & price, selling date & price etc.

Thanks

Jim


Some Real Estate APIs that may have the listing data, sales data you need:

Real Estate APIs to review

https://rapidapi.com/blog/best-real-estate-apis/

https://gist.github.com/patpohler/36c731113fd113418c0806f62cbb9e30

Is there an API to access MLS data? - Quora

	
Is there an API to access MLS data? - Quora








On Thursday, April 30, 2020, 3:19:14 PM EDT, SN Laabs <snlaabs@yahoo.com> wrote:


Hi Jim,

Finally. :) The software expert that I worked with for years and build our multi-million dollar software.

Summary:

After reviewing the code, he believes the model is likely accurate. 

But to know 100% accuracy of code: we need to test with data in the exact parameters he hard coded into the model - against the real time it took for the houses to sell. 

1. We are going to test the model with actual real estate information we pull (from a relator) to double check the accuracy.

2. Now, if the model is accurate off the data we pull, then we have a second issue, we can/need to resolve. Demarcation. 

    The value is hard coded in at 75k, for the value. It's not 'smart', so it's unlikely that the model would translate      to accuracy in price ranges of million dollars. So we would need to figure out a way to either fix that issue.
         Quick Brain Storming:
                 1. Drop down options for real estate agents to choose from in their local area, given the different price                     points.
                 2. Create a 'smart' AI component to calculate on the fly.  

Next Steps:

Find a real estate agent who can get us data based on the current hard coded data - run the code. Check the accuracy against what 'really' happened with the house sales.

***Biggest reason: Where is the working code that the university used to create the paper? My only guess: They don't have it. 

I never asked directly because I wanted to stay in good graces with the university, but this is a good explanation as to why suddenly the professor abruptly dismissed me after the student got back to him.  

Solution:

I am getting cost to see about writing new code to predict if this is even possible. (will let you know the cost when I get it)

Even if we own the code, even if some of it is open source, we have our own 'proprietary software', as you don't patent AI software.

Also with us communicating with the university we don't want a legal fight. In fact, their code being flawed actually may help us (see below why). 

University Solution: 

I wrote our attorney CW for confirmation, but I believe the most politically and legal aspect is to honor the agreement. We publish the universities code on our github (a code repository), and do what we said we would do. 

1. It meets our legal agreement.
2. It actually helps us (if our new code works), because most people who think they can copy the code and be direct competitors have no idea the paper is flawed and code is flawed. 
3. We don't embarrass the university and get into some sort of legal conflict.

*It is known the the university in earlier emails that I was creating our own code, but we found theirs more interesting. So I should easily be able to communicate that we decided to use our own code, but are sharing theirs per our license agreement.