GateScout is an AI-driven Hardware Design framework that explores whether reinforcement learning (RL) can synthesize gate-level digital circuits better than heuristic tools such as ABC. An RL agent builds each circuit one standard cell at a time, using cells from open-source libraries (SkyWater 130 nm, Nangate 45 nm, ASAP7). Every result is formally verified and measured for area, delay and power against ABC. With tabular Q-learning the agent already matches ABC on all 2-input functions, but it cannot scale beyond 3 inputs because its lookup table grows to tens of gigabytes. This project replaces the table with neural-network policies trained with deepRL on GPUs. The aim is to scale to larger circuits and libraries and to find where learned synthesis outperforms heuristic mapping.