Mubassher Ansari

I'm a M.Sc. Artificial Intelligence student at BTU Cottbus–Senftenberg. Before research, I spent three years working as a software engineer, and that experience left me with one persistent frustration — AI models that work beautifully in research often struggle the moment they meet real hardware. That gap is what brought me here. My thesis explores how to make neural networks run more efficiently on hardware accelerators by rethinking how precision is assigned across the network. The core idea is that not every part of a neural network needs the same level of precision — and treating them all the same wastes either accuracy or hardware resources. My broader interest lies in bridging the space between AI research and hardware deployment, particularly around quantization and how compressed models can be mapped intelligently onto real accelerator hardware. Outside of research, I work as a Werkstudent Platform Engineer at ADITUS GmbH, managing infrastructure for admission and event management systems — which keeps me grounded in real engineering problems. I am supervised by Mr. Mahdi Taheri and Dr.-Ing. habil. Christian Herglotz.