Workshop / Systems and experiments
A working index of hardware, infrastructure, robotics, and software projects.
Some are still changing, some run quietly every day, and some exist mainly because their failure taught me something worth keeping.
Active work where the implementation, setup, or operating model is still changing.
Learning design for additive manufacturing with a BambuLab A1, moving from basic Blender models toward parts that fit, support cleanly, and survive actual use.
Read build log →CAD fluency, support strategy, surface finish, tolerances, and material selection.
A maintainable soft-tube loop cooling a CPU and dual RTX 3090s across three 360 mm radiators. The loop is complete; thermal and acoustic tuning continue.
Radiator airflow, fan configuration, cable management, and reliable lighting control.
A Home Assistant installation on Raspberry Pi connecting lighting, climate, irrigation, and voice controls through local automation where practical.
Motion lighting, temperature-based AC control, automated plant watering, and voice actions.
Projects that have moved from setup work into regular operation.
A TrueNAS server with four 8 TB drives in RAID-Z2, used for family storage, media, and self-hosted services.
ZFS storage, ECC memory, containerized services, and a separate backup strategy.
A Quest 2 and wireless PCVR setup with a custom music library, competitive scoring, and custom sabers.
More than 200 custom tracks with stable wireless streaming and Scoresaber integration.
Robotics and networking projects from undergraduate and graduate research.
A mobile manipulation robot combining stereo vision, SLAM, autonomous navigation, and a robotic arm.
An autonomous robot that used vision and onboard sensors to locate and extinguish small flames.
A Linux congestion-control implementation designed to prioritize short flows in data centers.
Failures stay in the index when they changed how the next version gets built.
Three printed iterations looked plausible in CAD and failed in use. The hinge tolerances and PLA spring behavior were wrong for the mechanism.
What changed: prototype the mechanism in cardboard first, test fit before finish, and choose material from the load case.