What we actually do.

01

Search & answer engines

We take on a site and improve two related but genuinely different things: how well search engines rank it, and how reliably answer engines can quote it.

The second one is newer and widely misunderstood. AEO is not SEO with a new name. It rewards question-shaped headings, short self-contained answers near the top of a page, lists and tables that survive being extracted from their layout, declared entities, visible authorship and dates, and — most bluntly — whether your robots.txt lets AI crawlers in at all. A site can rank perfectly well and still be uncitable.

Evidence first, opinion second

We run this with an audit tool we built and maintain ourselves. It crawls every page of a site and scores it with a deterministic parser: each finding is a verified fact about the markup, not a judgement call. That evidence is then handed, identically, to four independent AI models, which argue about what it means and what to fix first.

Keeping those two halves separate is the whole point. Where all four models independently flag the same thing, it is almost certainly real. Where only one does, it is either an insight the others missed or an overreach — and you get to see which, instead of one blended voice telling you it is all urgent.

Three lenses on every page

• SEO — can search engines crawl, index, understand and rank this?
• AEO — can an answer engine lift a correct, attributable answer out of it and cite it?
• Accessibility & UX — can everyone actually use it?

02

Putting AI to work

A growing part of what we do is helping companies find where AI genuinely earns its place in how they operate — and, just as usefully, where it does not.

The hard part is rarely the technology. It is working out which parts of a business actually benefit, what it costs to run at volume, what happens when the model is wrong, and who notices. A tool that is right ninety per cent of the time is excellent for drafting and dangerous for billing, and telling those two cases apart is most of the work.

We start from your workflows, not from a demo

Where are the repetitive judgement calls? Where is information trapped in documents nobody has time to read? Where are people retyping things from one system into another? Those are the places worth looking, and they are usually unglamorous.

Then we build it — into the systems you already have, rather than as a separate thing to remember to use — and we build it so you can see it working: what went in, what came out, and what it cost.

Designed for verification, not trust

Our own site audit tool runs four different frontier models over identical evidence and shows you where they disagree, because a single model’s confident answer is not evidence. That habit is what we bring to your systems: keep the deterministic parts deterministic, make the model’s contribution visible, and put a person in the loop wherever being wrong is expensive.

• Finding the uses that pay for themselves — and saying so when there are none
• Getting your own documents and data into a form a model can actually use
• Building AI into existing workflows and software, not alongside them
• Keeping a human in the loop where a wrong answer is costly
• Running models on infrastructure you control, when the data cannot leave

03

Websites & CMS migration

A great deal of our website work is getting organisations off a legacy content management system they have outgrown — where upgrades break plugins, security patching never ends, and the content has been bent out of shape to fit somebody else's idea of a page.

We move that material onto something lighter that you control. Often that is our own site builder, built on Equipe — a PHP and MySQL framework we wrote and maintain in-house for building data-driven sites and applications. Sometimes it is something else entirely; the right answer depends on what your content actually is.

Content modelled to your material

Most CMS platforms give you pages and posts and expect everything to be one or the other. Real collections are rarely so obliging — an archive, a catalogue, a bibliography and a course listing each have their own shape. We model the structure around the material, so that it can be searched, filtered and cited properly rather than flattened into prose.

• Migration off legacy CMS platforms, content and structure intact
• Every old URL mapped and redirected, so accumulated ranking follows the content
• Structured content types built around your material
• Editing by the people who own the content, without needing us

04

Custom software & automation

Web-based applications built to your specification, for the case where nothing off the shelf fits the way you actually work — or where the thing that fits wants to hold your data hostage.

This is usually internal, unglamorous and load-bearing: the system a department runs on. Structured databases with controlled access, back-office and intranet tools, workflow and record-keeping systems, public front ends over private data.

Data flows, analytics and automation

The same work often arrives from the other direction: not “build us an application” but “these four systems do not talk to each other, and somebody spends every Monday morning reconciling them by hand.”

So we build the pipeline that moves and reconciles the data, the scheduled jobs that keep it current, and the reporting that turns it into something a person can act on — figures that are the same figures every time, rather than whatever last week’s spreadsheet happened to contain.

• Custom data pipelines between systems that were never meant to meet
• Reporting and analytics built on your own definitions, not a vendor’s
• Scheduled automation replacing recurring manual work
• Integrations with the tools you already pay for

Built to be left

We build on standard languages and standard databases, we document the setup, and your data stays exportable. A system you cannot leave is a system you did not really buy. We would rather keep clients because the work is good.

05

Servers & infrastructure

We specify, build, secure and maintain servers around what an organisation actually needs — primarily on FreeBSD and Linux. Not a preselected package you have to grow into, and not one you have already outgrown.

That ranges from a single well-configured application server, through fault-tolerant multi-server setups, to the day-to-day work that keeps them healthy: monitoring, backups, patching, DNS and certificates.

Research computing

At the larger end we have built Slurm-based parallel processing clusters for university research groups — scheduling scientific workloads across many machines, with the storage, queueing and user management that has to sit around them.

• FreeBSD and Linux server specification, build and hardening
• Fault-tolerant and multi-server configurations
• Slurm parallel computing clusters for research workloads
• Ongoing monitoring, backup, patching and maintenance

06

Hosting & email

We host the custom applications we build, and the websites we build or maintain. Email hosting comes with it, for domains we already look after.

The reason we bundle it this way is accountability. When the same people are responsible for the software and for the machine it runs on, nobody gets to point at anybody else when something breaks at 6pm.