About
The TANG Lab at the University of Pittsburgh is building the computing foundation for the agentic AI computing era. As AI models grow more capable, the challenge shifts to whether the underlying systems can train, serve, adapt, and reason at the scale, speed, and efficiency that future applications will demand. Our goal is to make AI infrastructure fundamentally more scalable, more efficient, and more accessible, from datacenter GPU clusters and heterogeneous accelerators to edge devices and emerging quantum platforms.
We approach this mission through a full-stack view of AI computing. Our research connects computer architecture, compilers, runtime systems, and machine learning systems to remove the bottlenecks that hold modern AI back: distributed memory and data movement across multi-GPU systems, irregularity in graph and foundation-model workloads, latency and energy limits on edge devices, and the software barriers that prevent new computing substrates from becoming practical. By co-designing across these layers, we aim to turn hardware complexity into usable performance.
Our long-term goal is to help define what AI computing systems should look like in the next decade. We envision platforms where large models can be trained and served with far less waste, where intelligence can move fluidly between cloud clusters and personal devices, and where quantum-classical systems can become practical accelerators rather than isolated experiments. Through this work, we aim to make the future AI computing stack not only faster, but also more adaptive, dependable, and capable of supporting ideas that today's systems cannot yet realize.