Technical director · CTO · Consulting
I lead engineering teams through work that has to land.
Statutory deadlines, regulated sectors, real users. I build delivery teams and keep them delivering – across cloud, web, mobile and SaaS product work. And from either side of the table, buyer or supplier, I advise on bidding, buying and holding suppliers to account.
The hard part is rarely the software.
Experience
Engineer first. Technical director since 2010. Delivered for HM Government, every UK mobile network operator, Landmark, Optasia and Petroleum Economist.
Where I come in
Before anything's built, or when someone's been told it's nearly finished, and has been for a while. My job is usually to find out what's actually true, then make it true.
When I leave
Not before it runs without me.
Where you come in
You don't need a brief. Tell me what's bothering you, and I'll say whether I can help.
Sectors served
- Government & public sector
- Financial services & fintech
- Telecoms
- Energy & critical infrastructure
- Legal
- Health & NHS
- Agritech & environment
- Industrial & connected devices
- Education
Centre-pivot irrigation, San Luis Valley, Colorado. The first time I checked a survey here, I thought the map tiles were broken.
Contains modified Copernicus Sentinel data 2026.
Selected work
Chosen for what they show, not when they happened. Named where I can, sector-only where I can't.
-
HM Government
Deadlines set by legislation, not by negotiation.
Two examples: a national register released for the date Parliament had fixed, and CefMAT, which standardised UK marine assessments and became a Copernicus Marine Service use case.
More detail Less detail
Statutory and document registers and their supporting services, for departments including Defra and MHCLG (then DCLG). In one case the launch date of a national register had already passed through Parliament before the build began, so the date could not move.
For Cefas, CefMAT: a web platform for answering marine policy questions about the waters around the UK, including assessing progress towards Good Environmental Status under the Marine Strategy Framework Directive and . It brought together large scientific datasets from very different sources – fish surveys, ferry boxes and Copernicus satellite and model data among them – across a wide range of indicators, from eutrophication, oxygen and temperature to the Large Fish Indicator. Long-running statistical queries ran in the background, producing tabular, graphical and map-based outputs in a standard format for national and international submission, with some indicators open to the public and others restricted to authorised users. Featured as a Copernicus Marine Service use case. A relationship maintained for roughly a decade, across several of Cefas's procurement cycles.
The mechanism by which fifteen governments and the European Union cooperate to protect the marine environment of the North-East Atlantic..NET · SQL Server · ArcGIS · Mapbox · Highcharts · third-party statistical libraries · Copernicus Marine Service data
-
Optasia
Every market its own operator, its own regulator, its own gateway.
Operational teams ran their own campaigns, to lists of up to ten million, in markets from India to South America, without waiting on engineering.
More detail Less detail
Their campaign management platform let operational teams run campaigns themselves: SMS and to subscriber lists of up to ten million, integrated with local messaging gateways at each operator data centre and deployed from India to South America, each market under its own operator and regulatory constraints. Connectivity couldn't be relied on, so messages were queued and retried until they got through – every bit as awkward as it sounds.
A real-time communication protocol used by mobile networks to let phones use simple interactive services (Unstructured Supplementary Service Data). For example, the menus you get when you dial a code such as *123#.A separate Handset Loan Portal let subscribers take handset finance through their mobile provider, funded by microfinance. Campaigns could lead into it.
Java · Postgres · ActiveMQ · Redis · Docker · SMS and USSD gateway integration
-
Field data capture
A survey has to show the surveyor was where they say they were.
CanopyCheck was adopted globally, including by PepsiCo.
More detail Less detail
Capture has to work offline and synchronise later, and what comes back has to be trustworthy: readings demonstrably taken in the field, which made "drive-by agronomy" much harder.
CanopyCheck, built for Landmark Information Group with NIAB and Cambridge University Farm, turned a phone photo into a ground-cover measurement. Tracked weekly through the season, it informed nitrogen, irrigation and yield decisions, and was adopted globally, including by PepsiCo. Crop4Sight, for Greenvale and Produce Solutions, carried canopy imaging forward: yield samples from Bluetooth scales, agroclimatology data from NASA POWER, and machine-learning models benchmarking each growth stage, early enough in the season for growers to act on it. Gaps in the NASA POWER feed were backfilled as data arrived, so predictions tightened quietly rather than visibly breaking.
Envirocheck Analysis for Landmark: most of the platform's initial API tier and the mobile app on top of it. The same offline-first pattern ran a defect management platform for an electrical maintenance company inspecting critical infrastructure.
iOS · Android · React Native · .NET · .NET Core · offline capture and deferred sync · machine learning · Azure · cloud APIs · NASA POWER
-
Wireless telemetry apps
A transmitter can broadcast to many devices or be paired for calibration, never both.
Used for monitoring loads held over people's heads.
More detail Less detail
Built native rather than cross-platform deliberately: the Bluetooth behaviour was the product, and an intermediate framework was a risk not worth carrying.
Users compose dashboards of live tiles driven by expressions they write themselves across multiple transmitters, with unit conversion handled automatically where the source types agree. The engineering sat in that exclusivity: configuring a transmitter has to propagate an offline state through every expression referencing it. No server and no accounts – everything on the device, projects shared as exported definitions, viewing and configuration each PIN-gated. One codebase, several brands.
Related work took sensor data somewhere with rather less tolerance for error: monitoring systems for power station control rooms, ingesting vibration data from turbine halls onto operator displays and handling exception notification.
Native iOS (Swift) · native Android (Java) · Bluetooth Low Energy · on-device storage · white-label builds
-
Online dispute resolution
Three roles, one dispute, and deadlines that carry legal weight.
The underlying rules engine survived every procedural change intact.
More detail Less detail
A claimant logs a dispute, the defendant accepts or rejects resolution, and an appointed decision maker reads both cases and issues a decision – three roles, each with its own view of the same dispute, with an administrator behind them appointing decision makers. Fees were taken through the platform at various points in the process.
The engineering was in time and locking. Acceptance expires after five days. Response windows run seven calendar days irrespective of weekends and bank holidays. Submissions lock when their window closes and only the decision maker can reopen one, for illness or absence. Every limit that looked like a constant – word counts, document sizes, the length of each window – turned out to be a policy somebody would want to change, so the rules became configuration rather than code. The procedure was revised repeatedly over the life of the project; the state machine underneath it did not have to be.
.NET · Angular · SQL Server · payment gateway integration
-
UK mobile short codes
Five competing operators, one authoritative register.
Run for six years across every UK mobile network operator, through a merger and a major rebuild. Still running.
More detail Less detail
It replaced a fragmented and error-prone process spread across five competing operators when the project began – O2, Vodafone, Orange, T-Mobile and Three – each with its own commercial sensitivities about what the others could see. Orange and T-Mobile later merged to form EE, leaving four.
Aggregators search and request codes; the controlling operator allocates. Reservations expire after three months, with an automated warning before they lapse, so the register stays honest about what is genuinely available.
I ran it for around six years, through the EE merger and a second major rebuild, then handed it on. It has been maintained and rebuilt since, but the core functionality is the same, and it is still running.
Also
Shorter, same point.
-
Motor trade
The same model of car can be a dozen different cars.
A trade-only marketplace, where subscription tiers decided who saw stock and when.
-
Luxury manufacturing
A catalogue too big for the App Store, demonstrated where there was no signal to fetch it.
Films, specs and 3D tours of every yacht, managed by the manufacturer's staff on the web and carried to sea on iPads.
-
Higher education
Sometimes the first deliverable is the specification.
A tender for a medical school system went back with suggestions instead of a price. The analysis and prototype that followed were what the whole build was procured against.
-
Publishing
Their product is the cartography. A map that worked but looked wrong was a map sent back.
PE's printed energy maps, rebuilt for the web to print standard. A separate World LNG Map app for iPad made its debut on a Perth trade show stand, in front of a key customer, on whatever Wi-Fi the hall had.
-
Health
The procedure you'll see twice a year is the one you can't get wrong.
A learning app for hospital staff: when a procedure was about to happen at a bedside, the staff who needed to see it could hear about it in time to be there.
-
Transport infrastructure
The people who knew the requirements had a bridge and a ferry to run.
A health and safety system for the Tamar Bridge and Torpoint Ferry, replacing paper and spreadsheets with one record, shared on a need-to-know basis.
-
Further education
Many integrations, priced as if they were all the simple kind.
A thin specification, a short deadline and a small fixed budget. Advised against.
Any of these makes a good first conversation. Pick one and ask me what the hard part turned out to be.
What I do
Delivery and recovery
A greenfield build, a team with a date it can't move, a build that has stopped telling the truth about itself.
Technical direction for an organisation not ready to hire it permanently. I start by finding out what's actually true: what works, what's been promised, and what the date really depends on. It usually depends on a team wider than the engineers: architects, testers, project managers, business analysts. Then I work out what the team and the business need, make it work, and hand it on.
Depending on the job, that means architecture, agreeing roadmaps, hiring, holding suppliers to account, or telling a board what it will cost and whether the schedule is real.
In agritech, that has meant services on deadlines set by the crop: data capture ready by planting, and machine-learning models ready by emergence to, for example, forecast yield and inform irrigation and burn-down decisions – every season, never quite the same twice.
Engagements run from a few days' assessment to a long-running relationship, sometimes across a portfolio of products.
Delivery is what I know best. The rest is the same judgement.
Early stage
Someone with an idea and nobody technical to ask.
Working out what's worth building first, and what isn't: a prototype small enough to be wrong cheaply, a plain answer on what doing it properly would cost, and a view on when to stop. Some ideas don't survive users, and it's better to find that out in weeks. The same approach applies to public sector teams who want something real in front of people before committing.
Buy or build
A buyer deciding what to commission.
Someone who knows their own field precisely wants a straight steer on whether the answer is SaaS, bespoke, something with AI in it, or a combination – before anything gets written down and sent out. Increasingly that includes whether AI is needed, and whether a supplier's claims about it hold up. My stance: show me.
Bid or no bid
A supplier with a tender in front of them.
Reading the for whether the document is coherent, whether the work is real, whether it is worth their time and mine, and whether there's a budget somebody will actually spend. Some get declined. Some go back, because the first job is getting the buyer to say what they meant.
Intelligent customer
A system bought, and nobody to hold the supplier to account.
The public sector's name for the buyer-side role that reads the change requests, questions the estimates, and says a date is slipping before the supplier does. I spent two decades on the supplier's side of the table – as both prime and subcontractor – which is most of why I'm useful on this one.
Expertise
A geospatial thread runs through much of my work – marine and environmental assessment, agronomy, energy assets – often building web front-ends because the established desktop GIS products' own browser offerings were not good enough.
- Cloud & platform (AWS/Azure)
- Distributed systems & architecture
- Legacy modernisation
- Greenfield builds
- Start-up delivery
- SaaS product delivery
- Project rescue
- Offline-first mobile (iOS/Android)
- Connected & industrial devices
- Geospatial & web mapping
- GIS
- Data & integration
- Government service standards
- PCI DSS & regulated delivery
- DevOps & CI/CD
- AI and machine learning: assessment and integration
I've built and shipped products with machine-learning models, so I know what they need before they're worth trusting: data that holds up, a decision they actually inform, and a way to tell when they're wrong.
Background
Technical director since 2010, independent since 2024.
2003–2007 · Software engineer, Digita (Thomson Reuters)
2007–2010 · Lead developer, Rokk Media
2010–2024 · Technical director and CTO, Rokk Media – mixed teams of up to 25
2024– · Consulting. Technical direction and advice to buyers and suppliers on bids, builds and contracts.
I started as an engineer and ended up running the delivery. Both jobs are still mine: the architecture conversation and the one in front of the board about what it costs and whether the date is real. I still write code when it's the fastest way to understand a problem, but I keep myself off the critical path: a team shouldn't stall because its technical director is in a meeting. Where a client will only deal with me, they get me. Bad news is mine to deliver. Good news belongs to the team.
Clients delivered for
At Rokk, Digita and independently.
- HM Government
- DCLG
- Defra
- Cefas
- Sport England
- St George's University Hospitals NHS Foundation Trust
- Tamar Crossings
- Microsoft
- HSBC
- NatWest
- Scottish Widows
- Arbuthnot Latham
- Cavendish Online
- National Grid
- Beran
- HP
- Petroleum Economist (Euromoney)
- Ashfords
- Screwfix
- Princess Yachts
- Mercedes-Benz South West
- Landmark Information Group
- Optasia (Channel VAS)
- Exeter Chiefs
- University of Exeter
- University of Plymouth
- SETsquared Partnership
- Every UK mobile network operator
Tell me what you're trying to get done.
Most conversations start with a problem someone can't yet name. If there's something worth looking at, we agree a first step together and decide from there. Some conversations go no further, some become a few days' work at an agreed price, some a longer engagement. I stay until it runs without me.
If it isn't something I can help with, I'll say so, and where I can I'll suggest who might. A first conversation commits you to nothing.
I usually reply within a working day, and always within three.
Tristan Stone
- tristan@tristanstone.co.uk
- linkedin.com/in/tristanstone
- 07743 246920
- Exeter, Devon · working UK-wide and remote