I want to build a Windows-only application that watches the live screen, detects people—including stylised in-game characters—in real time, and immediately drives a game controller so the crosshair locks onto and tracks that target as it moves. The workflow I picture is straightforward: the program captures frames, an AI model spots the person, and a virtual stick signal (XInput, vJoy or a comparable driver) nudges the aim every frame so it stays centred. Smoothness and speed are critical. On a 1080p feed I’m aiming for roughly 60 fps with no more than 40–50 ms end-to-end latency, so techniques such as YOLOv8, TensorRT or a lightweight custom network combined with OpenCV screen capture should fit. You’re free to choose Python, C++, or another language as long as the final build runs reliably on Windows 10/11 without needing exotic dependencies. Deliverables • Windows executable with a small GUI (start/stop, confidence slider, aim-speed slider, hotkey toggle) • Source code and the pretrained model that recognises people/in-game humans out of the box • Virtual-controller implementation that the OS and games read as a standard pad • Step-by-step setup notes covering driver installation, re-training the model, and build instructions Acceptance criteria • Consistently detects a single person on-screen and keeps the aim within ±5 pixels while they move • Holds 60 fps processing on a mid-range GPU (e.g., RTX 3060) • Runs for 30 minutes with no crashes or memory leaks If this sounds feasible to you, let’s talk about the best architecture and milestone plan so we can get a working prototype quickly and refine from there. I would also like the program to work with PS5 Remote Play, and therefore for the PC to rely mainly on the processor rather than the graphics card.