Roofline is a German AI infrastructure startup building a retargetable compiler that takes a trained AI model and deploys it across heterogeneous edge hardware — CPUs, MCUs, MPUs, GPUs, and NPUs — with a single Python call. Models from major frameworks such as PyTorch, TensorFlow, and ONNX are translated into optimised intermediate representations and then emitted as efficient executables for each target backend, including heterogeneous execution that splits a workload across whatever silicon a device contains.
The core problem Roofline addresses is the long-tail brittleness of edge AI deployment. Model engineers retrain models constantly, but porting them onto a new NPU, microcontroller, or accelerator typically means brittle hand-tuned toolchains, vendor SDK lock-in, and weeks of integration work. Roofline's layered compiler abstraction is designed to make those redeployments cheap and repeatable.
Roofline is a spin-off from RWTH Aachen University in Germany, emerging from the Institute of Communication Technologies and Embedded Systems (ICE) led by Prof. Rainer Leupers. The company was founded in 2024 by Maximilian Bartel, Thomas Zimmermann, and Moritz Joseph (CEO), with several additional ICE researchers on the founding team.
In July 2025 Roofline was selected for the European Innovation Council (EIC) Accelerator — one of 40 deep-tech startups picked from nearly 1,000 applicants — receiving €2.5M in grant funding plus a pre-committed EIC Fund equity investment for the company's next round. The award is one of Europe's most competitive deep-tech grants and is being used to expand the team and accelerate commercial deployments.
Roofline's differentiator is its academic depth in compiler engineering combined with a pragmatic, hardware-agnostic product surface. Its competition includes Apache TVM, MLIR-based stacks, and proprietary compiler tools from chip vendors themselves.