Show HN: I Turned DOOM into a Transformer

A developer created Doomformer, a project that compiles neural networks into DOOM maps and executes them within the ViZDoom engine using game mechanics like voodoo dolls and sector properties for computation. The system runs integer neural networks with perfect accuracy verified against NumPy, though it is extremely slow—approximately 23,000 times slower than NumPy—and takes 163 seconds of real-time DOOM gameplay to recognize a handwritten digit at default settings.