 Hi, this is going to be a very brief summary of typewriter, our approach for neural type prediction with search-based validation My name is Michael and this is joint work with George's Jason and Satish and we've all done this work while we were at Facebook As a motivating example consider a piece of Python code and nowadays there's a lot of Python code out there It doesn't really matter what this code is doing But the important bit here is that this code does not have any and some other code may have just a few Type annotations, so you for example do not know if you're statically reason about this code what parameter Type this parameter called color should have or what the return type of this function get colors actually is Now what you instead want to have are correct type annotations that are correct in the sense that if you add these annotations You program type checks, which means that everything is consistent now the approach that we are Presenting here is called typewriter and the key idea is that we combine two ways of getting these type annotations One is that we have a probabilistic type prediction model. It's a deep learning based model that looks at many pieces of information in your Python code and then Suggests likely types and then we are combining this with a gradual type checker that is type checking the code and We use this to perform a feedback guided search For type correct types among the predictions made by the probabilistic model Just to give you a few highlights of our results So the neural model that we are presenting outperforms the state of the art So it can predict more types and predict these types more precisely than any other type prediction model that has been Suggested earlier if you use this approach You can add 75 percent of the missing missing types While still guaranteeing that all the types you're adding are type correct So the approach will not bother developers with type incorrect suggestions And finally this tool is used at Facebook where it has been used to add thousands of otherwise missing types