Nikolaj Hindsbo
AI Engineer at Armada, Seattle
I'm an AI engineer working on the practical end of AI at the edge: getting open models to run well on real hardware, where accuracy, speed, memory and reliability all have to hold up at once. Most of my time goes to multimodal and vision-language systems, turning what those models can do into building blocks that agents, people and code can call reliably. I also work on time-series forecasting and AutoML.
I came to AI through engineering mechanics, drawn by an interest in space, evolution and how people learn. What I enjoy most is the messy part: finding what works, what doesn't, and what is practical enough to ship.
- Published SCOPE at HRI '26: an air-gapped camera agent built on separate planning, perception and control, with a 536-task sim-to-real benchmark in Blender. GitHub ↗ Paper ↗
- Run Armada's LLM and vision-language model work end to end: a catalog of open-source models from 0.8B to 1T parameters, hosted and optimized with vLLM and SGLang on hardware from datacenter GPUs down to briefcase-size edge modules.
- Built the pipeline that brings new open-weight vision-language models up to production standard, aligning prompts and output schemas to how each model was trained so agents can rely on the output.
- Benchmark models on our own hardware and data, and explain how choices like mixture-of-experts, speculative decoding and quantization play out for customers.
- Lead the AI work behind Atlas, Armada's operations platform, including time-series forecasting for Starlink and SD-WAN devices.
- Replaced reduced-order parts of full-engine simulations with LSTMs and neural networks, matching high-order accuracy while running faster.
- Cut a workflow repeated 20+ times a year from 200 engineer-hours to 20.
- Built and extended features in Ansys Mechanical across C++, C#, JavaScript and Python.
- GPA 3.97.
- TA for Using AI in Industrial Applications (49-734).
- GPA 3.86. Graduated with Distinction.
- VP of Badger Rail Club. Head Tutor at the Undergraduate Learning Center. TA for Statics.
- SCOPE — a camera agent and its digital twin — HRI '26, 2026
- Teaching an open model to read web pages — CMU, 2024
- Waste classification on the edge — CMU, 2024
- Soft robot legs that bend the right way — CMU, 2024
- SpacExxon: A spacecraft that refuels satellites — UW senior design, 2023
Python, TypeScript, MATLAB, C++, Java, JavaScript, SQL.
PyTorch, multimodal and vision-language models (Qwen-VL, Moondream, LLaVA), LoRA fine-tuning, quantization, mixture-of-experts, time-series forecasting, AutoML.
vLLM, SGLang, Ollama, MLflow, Databricks, Spark, AWS, Google Cloud, Git.
SolidWorks, ANSYS, Blender, Unity / VR.
Intro ML/AI, Deep Learning I & II, Multimodal ML, Generative AI, Trustworthy AI, Systems and Tool Chains for AI, CS Fundamentals, UX Development, VR, Multi-Variable and Vector Calculus, Linear Algebra and Differential Equations, Statistics, Applied Math Analysis.
Advanced Mechanics of Materials, Fracture Mechanics, Thermodynamics, Heat Transfer, Fluid Dynamics, Statics, Advanced Dynamics, Robotic Dynamic Analysis, Aerodynamics, Heterogeneous and Multiphase Materials, Tissue Mechanics, Advanced Materials Testing, Circuits, Advanced Controls Systems Integration.
Want a PDF copy? Get in touch and I'll send the current one.