Below is a rough write-up of our attempt at the admission puzzles for Hack MIT 2017.

URL: https://hotsinglebots.delorean.codes/u/USERNAME
Need to get the puzzler bot to like you. From the console, we know that puzzler bots only like automobiles. Our goal is to upload a profile pic that the neural network thinks is an automobile. In [username] file, it says, "Make /api/USERNAME/model/model.json, and model.hdf5 files private." When we add these to the base url, a model.json file and a model.hdf5 file are downloaded. The former contains a model of the neural network, whereas the latter contains the weights of the neural network. We will generate an image sych that the neural network thinks it is a car.

Install keras and tensorflow
Files are in allofkeras > keras > datasets
Start with a mutation rate of 0.1 and then slow decreases the number

python3 big_genetic.py
python3 image_converter.py

Upload output.png as profile pic
Like a puzzler bot and the date for the time machine will appear!

URL: https://captcha.delorean.codes/u/USERNAME/
The goal for the puzzle is to solve 10,000 out of 15,000 CAPTCHAs successfully. I solved 1000 captchas manually to be used as training and test data (manual_solver.py).

The lines are at the exact same position and angle for all the captchas for the same user. We can easily clean the captchas if we find the positions of the pixels of the lines (overlap of images).
Install opencv

python3 cleaner.py