Six things I built

The problem

An Indonesian retail investor placing a trade cannot easily answer three ordinary questions: what this trade costs once every broker fee is counted, what price the exchange's fraction rules will actually accept, and how far the auto-reject ceiling and floor sit from where the stock is trading today. Most people answer them with a mental estimate and a spreadsheet they update afterwards, if at all. The estimate is usually wrong in the same direction — optimistic.

What I built

A calculator and journal that computes fee-inclusive profit and loss, snaps prices to the correct exchange fraction, shows the ARA and ARB range, derives the true break-even, and imports and exports a versioned CSV so the record outlives the tool. The journal itself never leaves the browser — it is not something I want sitting on my server, and I would rather design that constraint in than promise it in a privacy policy.

The deployment is gated and token-verified. Error logs are scrubbed so that a stack trace cannot leak a ticker code or a price. Forty-eight tests exist for one purpose: to prove the system refuses things it should refuse — forged signatures, wrong audience, expired tokens, injected journal content, oversized payloads.

The problem

A doorbell draws almost nothing, but it has to draw it forever, and at the least convenient moment — after a week of overcast sky, at night. So the design question is not "will a solar panel run this", it is "what is the worst week this has to survive, and how much battery does that buy me". I sized the photovoltaic panel and the 18650 cell against that worst case rather than against an average.

What I built

The full CAD set — button housing, receiver enclosure for the PCB, solar panel housing, hinges — and the assembled unit around a solar charge controller, battery, speaker and transmitter pair. Then bench testing with a multimeter to confirm the charge behaviour matched what I had calculated on paper, which is a less satisfying step than it sounds and a more necessary one.

The problem

Bali received close to seven million foreign arrivals last year and tourism accounts for roughly a third of the province's economy, yet the performers who supply a great deal of what those visitors come to see have no infrastructure at all. Instagram is the discovery layer. WhatsApp is the contract. Cash is the payment processor. There is no registry of performers anywhere in the country, and middlemen take thirty to forty per cent of a fee for closing a gap that a directory would close for free.

What we designed

A three-sided marketplace connecting performers, host venues and private event clients, with the commercial model worked through properly: onboarding through the existing Banjar and Sekaa Gong networks rather than around them, a performer welfare fund, transparent dispute resolution — and, the part I argued hardest for, boundaries that performers set themselves. Some pieces are sacred and are not for a hotel lobby. The platform has to be able to encode that refusal, or it becomes another middleman wearing better clothes.

The problem

Commuters "blind board" — they queue where they happen to be standing, so a train leaves the platform with one carriage packed and the next half empty. Nobody is behaving badly; they simply cannot see the information that would change their decision, and the ten seconds in which it would be useful have already passed.

What we built

A camera in the cabin, AI detection to find and classify each passenger as sitting or standing, and a live 3D model of where the crowd actually is — projected as a hologram at the platform, colour-coded from empty through moderate to packed, so you choose your carriage before the doors open.

I owned the entire simulated build in CAD — the platform, the display unit, the train — along with the final product design and every poster and slide the team put up.

The problem

Today's quantum processors are noisy. Before anyone gets a fault-tolerant machine, the noise has to be characterised — you have to know what is going wrong and where it comes from before you can correct it in real time. That is a real research problem with a real paper behind it, and the case handed to us was blunt: this exists in a lab, make it a company.

What we built

A commercialisation strategy, not a prototype. A staged roadmap from TRL 3 to TRL 9 across the three hardware families that matter — superconducting, photonic and trapped-ion — with a separate development package for each, because they fail in different ways. A funding path benchmarked against real commitments in the sector, and a five-year financial plan ending in acquisition. Eight minutes to present it, five for questions, in front of an external mentor, the principal researcher and an industry judge.

The problem

In 2024 some 87,200 Singaporeans over 65 lived alone, a figure projected to reach 122,000 by 2030 — around one senior in four. Missed medication and missed appointments are not an inconvenience at that age, and the cases that make the news are the ones where somebody was found far too late. Yet every reminder app on the market assumes a smartphone, an account and a working connection, and the people who most need reminding are the least likely to have all three.

What we built

Two halves. A web platform on the doctor's side, where medication and appointment reminders are set and tracked remotely and tie back into nursing care and patient records — and a wearable buzz device on the patient's side, which alerts at the scheduled time and keeps alerting until the button is pressed, because a reminder you can sleep through is not a reminder. An ESP32 holds the schedule and drives the display and the alert; a serial bridge over USB carries the schedule across.