We build software
Full stack, from strategy to live product. Including AI that holds up in production.
An idea that needs building. A business that's outgrown what it runs on. We turn both into working technology.
What we've shipped
AI Research Engine
Grounded retrieval over a specialist corpus, with a citation for every claim and an evaluation harness behind it
Production AI SystemData Intelligence Platform
Unstructured public data turned into strategic signal, running autonomously on scheduled pipelines
Production AI System
Planiit
Invite and gifting platform with Stripe payments, gift cards, and live transactions
Web Application
Cotswold Planning Hub
Community resource for planning applications and local development
Community Platform
Capital & Country Homes
Land promotion, infrastructure delivery, and housebuilder
Website & CMS
FPL Ratings
Advanced data analytics platform for Fantasy Premier League
Data Platform
Yarnold Antiques
E-commerce and brand for curated vintage collectibles
E-Commerce
Maison Mama
Members' community for mothers in the Cotswolds
Membership PlatformCapital Land Property Group
Security-first replacement of an end-of-life platform for a strategic land developer
Platform RebuildThe demo is easy.
Production is the work.
Any model can answer a question about a document you hand it. The engineering is holding that answer correct across a corpus that keeps changing underneath you, catching the query where retrieval quietly missed something, and proving it still works after the next model upgrade. Every system below was built the same way: we're the AI expert, you're the domain expert.
Nobody reads the whole archive. The work is detecting the ones that changed overnight and making sure nothing still cites the clause that was withdrawn.
Agents that recover. The first plan failed on stock. It replanned, then stopped at the point a human should decide.
A better model is not always a safer one. The upgrade improved retrieval and started guessing where the old one refused, so it does not ship.
Four stages.
One ongoing partnership.
Understand
We embed in your business like an in-house CTO. How does it actually operate? What's holding it back? Where can technology create a competitive advantage? Proper discovery, not a questionnaire.
Plan
A technology strategy with real costs and timelines. What to build first, what to leave for later, and why. A decision-ready roadmap, not a 60-page strategy deck.
Build
Web applications, AI systems, platforms, or connecting systems you already have. We design and engineer the software in-house. No outsourcing, no handoffs.
Partner
Ongoing technology leadership for those who want it. We host, monitor, and evolve your stack as the business scales. Quarterly reviews, proactive recommendations, and continuous access to the engineering that built it.
The technology behind
the work
We choose the right tool for the job. These are the ones we reach for most.
Software development and technology strategy
for businesses at every stage.
Established businesses ready to modernise
You know technology should be doing more for your business. Internal tools that don't talk to each other, manual processes that should be automated, data you're not using. We come in as your outsourced technology function, assess everything, and build a clear path forward.
Startups and new ventures
You need what a technical co-founder brings, without the search. Business model, go-to-market, product strategy, and the engineering to make it real. We've built our own products from scratch. We do the same for clients at every stage from idea to live software.
Perspectives on technology,
product, and business
Most of what gets sold as an AI agent should be one API call
Anthropic's own guidance says start with the simplest thing that works. Four levels of complexity, what each one costs you, and how to tell which one your problem actually needs.
Approval is not a security control
Anthropic's telemetry found users approved roughly 93% of permission prompts. Why 'you approve each action' is a liability transfer rather than a safeguard, and what to ask for instead.
The definition of done should be a number, agreed before anyone builds
Most AI projects are accepted on a demo and a feeling. What changes when the acceptance criterion is a score on a test set both sides signed off in advance.
Let's talk about what
your business actually needs
Tell us what you need. We'll work out how to get there.