Nick Oak blog
← Back

Hardware Lab

The physical-world bench behind the software-heavy work.

This is not one polished product. It is the hands-on thread where factory AI, spatial computing, and personal automation stop being diagrams and start touching plastic, cameras, sensors, GPUs, and rooms.

The first branch was 3D printing. Around 15kg of Bambu A1 printing experience had accumulated across memory, ChatGPT chats, OrcaSlicer profiles, failed parts, and successful PETG/PLA/TPU runs. I turned that mess into a knowledge base: material notes, temperature and cooling tables, failure-fix matrices, and reusable settings instead of relearning the same lesson after every print.

The second branch was Jetson Orin Nano. The useful receipt: JetPack 6.2, NVMe boot, Docker and jetson-containers working, NanoOWL running at roughly 30 FPS on real video, and a web UI visible from the Mac. Not a benchmark flex. A sanity check that edge AI on small hardware is real enough to reason about for factories, robotics, and physical interfaces.

The current branch is apartment splat capture: phone video into selected frames, COLMAP poses, Brush training, and local inspection in a reused Spark / Three.js viewer. The first useful artifact is a good room splat. The second is a private Mac Mini pipeline where I can process a space without sending the apartment to a cloud service.

This also loops back into Tortuise and the broader Stark Splats direction. Terminal splats are one weird surface. Apartment capture is another. Jetson edge vision is another. The common thread is simple: if agents are going to operate in the real world, I need working intuition for the hardware edge, not just API calls from a comfortable chair.

Status as of May 2026: the 3D printing knowledge already paid off in better print settings and visibly improved quality, especially for PETG Translucent. Jetson onboarding is graduated, apartment splat capture is active, and most of this stays private until there is a clean demo or a useful public write-up.

Related: Tortuise, Factory AI Consulting, and Gaussian Moat Cuda.