I'm a third-year Mechanical Engineering student at the University of Toronto. I build end-to-end electromechanical systems.
I believe in first-principles empiricism: derive constraints from physics, then expand knowns through controlled iteration. This helps me design and solve problems in a grounded and quick fashion.
Machined a copper condenser for a wickless heat pipe and tested it charged and dry: 0.074 K/W against 0.26–0.28, steady at 410 W.
Still open: The centre ran 18 °C above the edge, and that gradient per watt was the same charged and dry.
Recovered the unique 121-bit key from the chip's layout and one sample waveform, then showed the chip is an 11×11 Star Battle checker.
Part 2: What the chip computes
What it prints: (* TWO STARS *) for the key. All zeros prints EMPTY SKY, all ones BIG BANG, everything else TRY AGAIN.
A 3D-printed record that plays. v2 writes the groove as G-code and round-trips a 440 Hz tone to 100.5 µm against 100 commanded.
Wrong turn: v1 cut depth because a 0.48 mm bead seemed too wide to wiggle. Written as G-code directly, it places to about 0.01 mm.
Modelled the chapter's existing furnace before measuring it: a ±5 °C flat zone of 26 mm (17 to 41 over the unmeasured inputs), 48 s to the quartz limit if the relay fails closed, and an element at 1153 °C at full power.
An optimizer-sized 3D-printed 3:1 gearbox, exact 3:1 with gears at most 30 mm across. A hand check later found its safety factors wrong.
Requirements and CAD for a 4-DOF camera robot: 17 verifiable specs, $1,435 BOM, 28% under budget. Not built.
A printed compliant joint for a rehab glove: elastic to 90° in nonlinear FEA, restoring torque above the 0.012–0.015 N·m friction floor.
A tremor-damping glove built in 48 h: Best Prototype and Top 3 at NeuroHack 2025. Tremor reduction not quantified.
