Selected work

Where the systems actually ran.

Not a portfolio. A short record of deployments in conditions that do not forgive — national networks, field programmes, regulated supply chains — because a claim about engineering discipline is worth nothing without somewhere it was tested.

20Years building
100K+Households reached
~65Engineers
6Sectors
TelecomGrameenphone

Biometric SIM verification at national scale

Identity verification for SIM registration across a national mobile network, operating at the volume and uptime a carrier requires.

What made it hard

Biometric capture on handsets that cost less than a day's wages, over connectivity that drops mid-transaction, at a throughput where a one-second regression becomes a queue in every retail outlet in the country. Failure is not a bad user experience — it is a person who cannot get a phone number.

What it shows

Identity systems under real load, and engineering for the device people actually own.

DevelopmentBRAC

Field delivery verification, ultra-poor graduation

Verifying that assistance reached the household it was intended for, across a programme operating at a scale of hundreds of thousands of households.

What made it hard

The user is a field officer in a village, not an analyst at a desk. The network is absent for most of the working day, so capture must be offline, signed at the point of delivery, and reconcilable later without creating a second version of the truth. A donor asking what happened must get the same answer months afterwards.

What it shows

Offline-first evidence capture, and the discipline that became the grade ladder.

DevelopmentTrickle Up

Last-mile programme monitoring

Monitoring and evaluation for a US non-profit running graduation programming, with field data flowing from the point of delivery into donor reporting.

What made it hard

Donor reporting and field reality rarely use the same vocabulary. The system has to serve a supervisor who needs today's visits and a funder who needs a defensible annual figure, from one record, without either side editing the other's view of it.

What it shows

One record serving two audiences with different incentives — the same problem a passport solves between factory and buyer.

IndustrialNational agro-input network

Fertiliser authenticity and diversion detection

Detecting counterfeit product and distribution diversion across a national agricultural input network.

What made it hard

Product leaves the plant genuine and does not always arrive that way. Finding where a chain is being gamed means inference over scan and movement data that was never collected for that purpose, in a setting where being wrong about a distributor has consequences for a real business.

What it shows

The fraud-detection work now the subject of a US National Science Foundation research proposal.

Supply chainBangladesh apparel exporters

Digital product passports, ahead of the 2028 regime

Issuing product passports with exporters in Bangladesh, built to the CEN standards cited in the Official Journal in July 2026.

What made it hard

The data a passport needs is produced on a factory floor by people with no reason to care about a European regulation. Collecting it as a separate compliance exercise fails within a quarter. It only holds if it is captured as a by-product of running the business.

What it shows

The product line this became, and why the ERP sits underneath it rather than beside it.

LearningDevelopment agency programme

Immersive training for industrial skills

A virtual-reality learning build for an international development programme, delivering practical skills training where the equipment, the site or the risk makes real-world practice impractical.

What made it hard

Immersive learning fails on hardware constraints long before it fails on content. The build has to hold frame rate on accessible headsets, be usable by someone who has never worn one, and produce a record of what a trainee actually completed rather than what they were shown.

What it shows

Range beyond our product lines, and the same instinct — a completion record that can be checked, not asserted.

How this gets delivered

Silicon Valley leadership, Bangladesh delivery.

Much of this work is delivered with BMQAS, our Bangladesh delivery partner, under SurroundApps technical leadership. Architecture, security review and the evidence model are ours; the delivery capacity that makes a national rollout affordable is theirs. It is the arrangement that lets a sixty-five-person company take on work normally priced for a firm five times the size — and it is why our engineers have field experience that firms selling from a US office do not.

What connects them

Different sectors. The same question.

A carrier verifying a subscriber, a donor verifying a delivery, a buyer verifying a supplier, a trainer verifying a completion. In every one, the person who needs the assurance is not the person holding the evidence. That is the problem we build for, and these are the conditions we learned it in.