Chess is a classic testbed for AI research, offering complex strategic and tactical challenges. In the modern day, the most commonly used chess bots are ones that rely on predefined strategies or expertise of humans relying on a brute-force method to come to the right decision. This paper explores the advantages and possibilities in creating a chess bot that can learn autonomously, through a repeated reward and punishment system implemented through reinforcement learning. The usage of reinforcement learning allows the bot to move in more complex format and react to the movement of its opponent intuitively. This is being achieved through the usage of Monte Carlo Tree Search (MCTS) integrated with a neural network. This paper aims to use the advantages reinforcement learning and the nuances of game theory to develop an intelligent self-learning chess bot. The proposed system involves two major steps: self-play to generate training data and continuous model updates using reinforcement learning. The chess board states and game outcomes from self-play act as the dataset, helping the bot refine its move selection over time. The evaluation metric is the bot’s ELO rating, measuring its performance improvement against top-level chess engines and tracking its progression to superhuman play. The bot’s estimated Elo rating increased from 1200 to 1350 over the span of five thousand games played independently. Subsequently, the bot’s success rate increased to 65 from 25%.