Work in progress ...
I'm currently working on some personal projects.
I will publish them here when they are ready for release.
Pokemon Battle Simulator Engine
A lightweight, modular Pokemon battle simulation engine built in TypeScript. This project provides a clean architecture for simulating turn-based Pokemon battles, featuring a flexible type system, move mechanics, and battle state management.
Designed for extensibility, it serves as a foundation for building Pokemon battle games, AI training environments, or competitive analysis tools. The engine handles core battle logic including damage calculation, type effectiveness, status effects, and turn order resolution, all wrapped in a developer-friendly API.
ParoleNet
ParoleNet is a multimodal model for predicting turn-taking. Its main objective is to determine, at the end of a given sentence, whether the person currently speaking will continue to speak or yield the floor to their interlocutor. It takes as input the last 20 words of the sentence as well as the final two seconds of the audio recording and performs classification on two turn-taking classes.
ParoleNet has been trained on a specific dataset of the French language containing 16,400 sentences. We have observed encouraging results, where the model has learned to anticipate turn-taking with performance superior to a randomly initialized model. Analysis of the statistical results suggests that the model is capable of understanding a relationship between the input data and the prediction of turn-taking, rather than merely a statistical prediction based on the frequency of each class uncorrelated with the input data. These observations indicate significant learning by the model, thereby reinforcing the validity of our approach.