Nik Bamert

ML Systems Engineer based in Zurich, Switzerland.

I build efficient ML systems where algorithms, infrastructure, and production constraints meet. My background spans embedded systems, computer vision, quantization, retrieval/search, and production ML infrastructure. I care about the boundary where algorithms meet systems constraints: latency, memory, reliability and performance in real-world systems.

You can reach me through LinkedIn or GitHub.

Selected work & background

Independent — ML Systems Engineer · 2025–present

Building efficient ML systems through benchmarked work on search, optimization, inference, and performance-critical infrastructure. Current work includes semantic search over scientific corpora, automated LLM-driven kernel optimization, and cache-conscious C++ data structures.

→ Building a Lightweight Semantic Search Engine for arXiv with Fine-Tuned Embeddings (Apr 2026)
→ AgenticKernel: 43× C++ Kernel Speedup with LLMs (Apr 2026)
→ A Cache-Conscious Hash Join for 19× Faster Binary Descriptor Matching (Mar 2026)

Pallon (ETH spin-off) — Software Engineer, ML / CV Systems · 2020–2025

Built and operated production CV/ML systems across data pipelines, 3D reconstruction, model development, deployment workflows, and reliability.

Carried primary technical responsibility for production operations of the CV pipeline, spanning data ingress, processing, reconstruction, and inference. Worked closely with frontend, product, and customer support teams as the system scaled to millions of images per day for infrastructure customers across Europe and North America.

ETH Zurich, CVG — Research Engineer · 2019–2020

Worked on neural network compression, quantization, and efficient representations with custom CUDA/C++ kernels.

ETH Zurich — BSc & MSc, Computational Science and Engineering · 2014–2019

Thesis work focused on efficient representations for machine learning and vision systems, including quantization-aware training, binary neural networks, and fast retrieval and matching of binary features.

→ SIMD implementation of PatchMatch stereo (2019)
→ fade - fast adaptive stereo estimation (2018)
→ openGPC - efficient sparse binary feature matching / depth estimation (2017)

Independent Consulting — Embedded Software Engineer · 2007–2014

Built bare-metal firmware, bootloaders, and custom drivers for real-time systems under tight memory and compute constraints for clients in the public transport industry. Software deployed in public transit infrastructure across Switzerland, wider Europe, and the US, including Zurich.