In recent decades, tensor networks have become powerful tools for studying complex many-body systems, including strongly-correlated quantum states at low-energies and finite temperatures, classical partition functions, and highly disordered systems such as spin glasses. Matrix Product State (MPS) is a pioneering and highly effective tensor network for simulating one-dimensional quantum systems. Its extension, Projected Entangled Pair States (PEPS), tackles higher-dimensional systems and has advanced the understanding of interesting physical systems such as quantum spin liquids. While MPS and PEPS excel on regular, low-dimensional lattices, many important models, such as spin glasses and quantum chemical systems, are defined on irregular and densely connected graphs. In this talk, I’ll describe how the PEPS algorithm can be adapted to such graphs. A key aspect of our algorithm is that it allows the geometry of the tensor network to dynamically adjust to fit the specific correlation structure in the problem. I will present some benchmarking results from simulations of spin glasses, simulated quantum annealing, disordered spin systems, and also from simulations of regular 2D lattice systems, including the recent IBM kicked Ising experiment on the heavy-hexagon lattice [1]. [1] Efficient tensor network simulation of IBM's largest quantum processors, Phys. Rev. Research 6, 013326 (2024)