Vadim Markovtsev

The talk will demonstrate how doing less can result in doing more: how domain-specific data constraints may lead to dramatic speedups in serialization. There will be two examples: binary caching of numpy object arrays (Pandas) and generating json web responses from @dataclass-es. We will also cover the revolution of using C++17 in Cython, custom arena memory allocators based on mimalloc, and gory internals of CPython.

Many developers hate doing code reviews. Reading foreign code is hard, and suggesting improvements is even harder. Yet a dramatic portion of code review time goes to figuring out the boring details: formatting, naming, microoptimizations and best practices. We believe that all of those can be automated with ML on Code, either learning from a particular project or from all the open source code in the world which is relevant. This talk will be about open source "analyzers" - ML-driven code review agents which deal with the boring but important details.