Undergraduate in Malaysia. working on quantum optimal control at the pulse level, and on the classical electronics sitting between the optimizer and the qubit.

Merged upstream

• harmoniqs/Piccolo.jl#238 — dual-rail encoding and EmbeddedOperator utilities: sparse subspace reduction, logical-ordered subspaces, operator/target consistency checks, 10 testitems. unitaryHACK 2026.
• harmoniqs/Legato.jl#27 — FFT pulse-spectrum analyzer. unitaryHACK 2026.

My own

• predistort — a differentiable model of a superconducting control line (DAC zero-order hold, line response, IQ imbalance, amplifier compression) wired to a three-level transmon in JAX, with every stage adjoint derived by hand.

  The composed Julia↔JAX gradient agrees with central finite differences to 1.09e-06 worst-case, and make env && make build && make verify reproduces that from a clean clone (verified on Apple silicon).

  The headline result is negative. Below roughly 131 MHz of drive bandwidth the linear inverse demands DAC codes outside the box, and a clipped inverse is not an inverse — the corrected gate (3.76e-03) comes out worse than no pre-distortion at all (2.40e-03).

I use AI assistance when I write code, and I disclose it where the work is published. Nothing goes out that I have not read, run and tested myself.

I came to this from embedded and storage systems: STM32/FreeRTOS firmware, an automotive radar project, and a ZNS emulation result (1st runner-up, Sandisk symposium).