1) A NoSQL database best suited our project. So, the non-nativeness was a hurdle. Neo4j bindings for Python used JPype, which was rather unmaintained. And I read at some places that it was a bad idea to mix python-java (especially using the unmaintained JPype).
2) Using Flask gave us so much flexibility as a framework. SQLAlchemy was probably the best thing that I've ever seen. But the work was made complex by the fact that Python was not able to give us the performance that we wanted. We tested the web-project running locally, and we noticed some exciting results. Even though we had done proper memory management from our side, the memory consumption was a little high than we had actually expected. I'm not an expert web-technologies, but as a programmer for last n years, I've learnt a lot about memory management. We projected that when the load would be actual, python performance would suffer.
3) We wanted a real-time performance application so that the user experience was the best. We implemented a part of the web-application and tested it. The results were unsatisfactory. Then we planned on shifting to scala.