Building is the fastest way to learn.
A glimpse into the research directions I'm actively exploring. From accelerating LLMs to building multi-modal intelligent agents. This is what's coming next.
How I extended the RSL-RL library to support multi-discrete action spaces, enabling new classes of optimization problems beyond continuous control. This post shares the motivation, implementation details, and benchmark results compared to Stable Baselines 3.
Using Reinforcement Learning to coordinate multiple kinetic effectors and defend a sensitive area from large-scale kamikaze drone swarms, trained entirely in a custom-built, high-fidelity simulation.
A platform for training, submitting, and evaluating DeepRL agents in arcade-style environments with real-time tournaments, leaderboards, and community-driven competition.
