Things I've built
Machines, the electronics that run them, and the software that makes them useful.
LOOM
3D Printing with Embedded Electronics
A desktop machine that prints a plastic part and builds working circuitry inside it — in one job, with no one touching it halfway through.
Four tools ride a single platform and swap automatically: an FDM hot end, a pneumatic paste dispenser, a vacuum pick-and-place head, and a probe with a camera. Most of the difficulty is not mechanical — it is getting one job to interleave printing, dispensing and placement.
BAJA SAE Electronics
Three Boards, One Race Car
I lead the electrical and electronics subsystem of AUB's student-built off-road race car — three ESP32s across three boards, covering data acquisition, the driver dashboard, and telemetry back to the pit over LoRa.
Power is LiFePO4 and an alternator, sized for 4+ hour endurance runs. Over 40 sensor channels stream at 20 Hz into a Python desktop client with live GPS mapping and post-race lap playback.
The RedRoom PMS
Print Management for the AUB Makerspace
The tool the engineering student body uses to get things printed. It runs inside AUB's makerspace page as an authenticated iframe — not a demo, a thing people use.
Staff work print queues, inventory, analytics and access control across seven permission-gated pages; students track their own jobs through a separate portal. Nothing renders until the session has been checked.
PCS
Printer Control System
The half of the platform that touches physical machines. Staff send a sliced G-code file, then pause, resume or cancel it — from a browser tab.
A Cloudflare Worker holds every secret server-side, so none reaches the browser. Behind it, a FastAPI relay on a Raspberry Pi 5 drives three Klipper printers over the Red Room's own 4G link. Their IP addresses drift on every power cycle, so the relay identifies each machine by MAC address and re-finds it on its own.
3DPME
3D Print Material Estimator
Filament weight estimation without running a slicer — so the portal can quote a job the moment a student uploads it.
It reads the mesh itself: convexity, bounding-box fill, surface-to-volume and height ratio classify each STL, ZIP or G-code upload. Mean error is 3.14 g across 52 real prints.
10% infill