roofline Tech Stack

Compiler platform for deploying AI models across edge hardware

Software Development Cologne 11–50 employees Founded 2024 Privately Held

Roofline builds a compiler-first platform for edge AI deployment, built on MLIR and LLVM with deep support for PyTorch, TensorFlow, and ONNX. The tech stack and active projects reveal a compiler-heavy engineering org focused on optimization primitives (vector instructions, tiling strategies, quantization flows) rather than a higher-level inference framework — positioning them as infrastructure for hardware diversity (RISC-V, CUDA, Vulkan, NPU) rather than a single-target solution. Early-stage hiring (intern-heavy, 5 engineers total) and open-source partnerships suggest a research-to-production arc.

Tech Stack 17 technologies

Core StackC++ Python PyTorch TensorFlow Linux MLIR LLVM RISC-V Jax ONNX ExecuTorch TensorFlow Lite C Vulkan CUDA OpenCL Git

What roofline Is Building

◆Challenges

Improving benchmarking infrastructure
Enhancing compiler reliability
Detecting accuracy issues

▲Active Projects

Partner with open-source community
Next generation compiler build upon mlir
Explore novel optimization techniques
Mlir interpreter
Optimizing performance with vector instructions for risc-v
Blockbuster (new tiling strategy)
Edge-ai model quantization flows
Quantization flow for compiler
Extend host compiler and runtime
Coordinate heterogeneous cpu/gpu/npu execution

Hiring Activity

Minimal5 roles · 0 in 30d

Department

Engineering

Seniority

Intern

Junior

Mid

Senior

Company intelligence

Find more companies like roofline by tech stack, pain points and active projects

Get started free

About roofline

Roofline solves fragmentation in edge AI deployment by abstracting away hardware complexity. Users import trained models (PyTorch, TensorFlow, JAX, ONNX) and compile them once to run across CPUs, GPUs, NPUs, and custom accelerators with a single Python interface. The company is based in Cologne, Germany, and focuses on compiler infrastructure, quantization flows, and heterogeneous execution orchestration. Current engineering efforts center on improving compiler reliability, building benchmarking infrastructure to validate accuracy, and developing novel optimization techniques like their Blockbuster tiling strategy.

HeadquartersCologne

Company Size11–50 employees

Founded2024

Hiring MarketsGermany

Frequently Asked Questions

What frameworks does Roofline support?

PyTorch, TensorFlow, JAX, and ONNX. Models are imported into the compiler, then deployed across target hardware via a unified Python interface.

What hardware does Roofline compile to?

CPUs, GPUs (CUDA, Vulkan), NPUs, and RISC-V architectures. The platform abstracts hardware diversity through MLIR and LLVM-based compilation.