A simple letter/digit recognition neural network

I need a complete and executable pyhon code for Tenserflow letter recognition neural network, code should be executable from any computer using Tenserflow framework. Dataset: For data set I need 10 000 letters/digits(even distribution) from NIST Special Database 19 ([login to view URL]). Creating a final dataset of 9 000 letter/digit Training dataset and 1000 letter/digit testing dataset. As far as I see there is no real need for any validation dataset. Tell me if I am wrong. Other limitations: Tenserflow neural network should contain at least a few Tenserboard diagrams (like accuracy, cross entropy loss, maybe weights etc). Similar example for reference can be seen here: [login to view URL] But I need just a basic running model. It does not have to be pretty or advanced in any way. Here are deliverable's I need. 1) Python code which will work for Tenserflow and will "read/load" NIST SD 19 format dataset (128x128 px grey-scale images). 2) Python code with simple but complete neural network (convolution neural network will be a big plus, but not mandatory): Reading dataset, network training, network testing (while displaying accuracy for the iterations(epoch) set by user(in code)) 3) Previous two points + a few Tenserboard diagrams (like accuracy, cross entropy loss, maybe weights etc). 4) A simple network optimization, but not necessary. The time frame is really tight!