This document presents Wi-Counter, a smartphone-based people counting system that leverages ubiquitous Wi-Fi infrastructure. It consists of three phases: a crowdsourcing phase where a mobile app collects Wi-Fi signals and people counts; an offline training phase where the data is preprocessed and used to train a neural network model; and an online counting phase where the trained model estimates people counts based on new Wi-Fi data. An experiment showed Wi-Counter can reliably count people with up to 93% accuracy by modeling the complex relationship between Wi-Fi signals and number of individuals present. Future work will explore using this approach for people counting over larger areas.