This paper presents a model-free approach to optimal trade execution using Double Deep Q-Learning (DDQN). The authors develop a neural network that takes state features (time, inventory, price, quadratic variation) and an action as input and outputs the Q-value. The network is trained using experience replay and double DQN to mitigate overestimation and instability. The reward function captures profit from price changes minus a quadratic penalty for market impact. The terminal inventory constraint is handled by adding a final liquidation step. Applied to nine stocks, the DDQN strategy outperforms the time-weighted average price (TWAP) benchmark in terms of mean/median outperformance, probability of outperformance, and gain-loss ratio, demonstrating the effectiveness of reinforcement learning for execution without relying on parametric price models.