Welcome to the land of graphs, your English is great.
I think your requirements for your system are very reasonable and shouldn't pose a problem. Sounds like a good cause.
You can also use python against Neo4j-server which should help you get going faster (e.g. py2neo) there are also embedded jvm-python bindings.
As resources, have a look at the videos a
video.neo4j.org and the manual at
docs.neo4j.org
Regarding your memory requirements, that's not that much, we are running the test-plans on heroku with a similar size, they work but are not most performant and for some queries the memory is not enough to hold intermediate results.
There we also have the approach of hosting many databases on a single machine (hosted on AWS) where we even suspend databases that were not in use for a while and restart them on the first http-request, which saves a lot of resources.
I'm not so sure about existing applications in that space, Thomas Baum wrote a CRM like app using Neo4j, Java, GWT
In general grab a whiteboard and draw the elements of your domain, that becomes your graph and then try to work from your individual use-cases.
I think before wondering about deployments you should work on getting a proof of concept done and see how well you can model your domain and focus on the integration of your webapp with the graph database.
After that we can still discuss the deployment, I think cloud deployment of the graphdb instances is also an option (e.g. on heroku).
Cheers
Michael