My name is Auke. I'm working on robots for autonomous construction at Monumental, based in Amsterdam.

I focus on the computer vision stack (Python), efficient localization and planning (Rust), and accessible interfaces for robot operators (Typescript). We're building houses! If you're ambitious, don't mind getting your hands dirty, and want to build something tangible, you should join us.

I'm also interested in running open-weight models on consumer-grade hardware and edge devices. For example, it's already possible to run medium-sized language models on Apple silicon.

I was previously a research scientist/engineer at Qualcomm AI Research, where I was fortunate enough to publish on multiple topics, including efficient autoregressive inference, neural data compression, on-device deployment of neural models, and synthetic data generation.

Before Qualcomm, I trained neural networks using Theano as data scientist at Scyfer (acquired by Qualcomm in 2017).

Public talks, workshops, open-source work

• Producing adversarial audio.
  How to trick OpenAI's Whisper into producing wrong transcriptions.
• mlx_transformers_benchmark
  Benchmarking local LLMs and mlx operators on Apple silicon.
• Outstanding reviewers CVPR 2025.
  Nice to be acknowledged (5.3% of reviewers).
• Efficient Large Vision Models workshop @ CVPR 2025.
  Renewed! Co-organizing.
• Efficient Large Vision Models workshop @ CVPR 2024.
  Co-organizer.
• TWIML AI podcast
  I was a guest at Sam Charrington's excellent ML podcast.
• Efficient transformers: towards deployment on edge devices.
  Invited talk @ Transformers at Work 2022.
• Efficient Neural Codecs.
  Invited talk @ CLIC 2022 (CVPR workshop).

Selected publications