What we build
Five kinds of work
Most projects start as one of these. In practice they overlap: an agent usually needs automation underneath it, and analysis is often what makes either one worth having.
The interfaces and examples on this page are drawn by hand to make each category concrete. They illustrate the kind of thing a project might involve. They are not product screenshots, and not a portfolio of delivered work.
AI agents
Software that carries out multi-step work on your behalf, not just one prompt at a time.
An agent is software that can take an objective, work out the steps, use the tools it has been given, and keep going until the job is done, checking its own work along the way.
The engineering that matters here is scope and safety: deciding exactly what an agent is allowed to touch, what it must hand back to a person, and how you can see what it did and why.
What it can look like
- Reading an inbox and drafting replies for a person to approve
- Researching a topic across sources and returning a referenced summary
- Working through a queue of tickets, resolving the routine ones
- Preparing a recurring report by gathering the inputs itself
Illustrations of what this covers, not a list of completed projects.
Voice AI agents
Agents you talk to rather than type to, built for practice: music and language training.
A voice agent listens, answers, and keeps a conversation going in real time. That suits practice better than a text box does, because the thing being practised is itself spoken or played, and the feedback has to arrive while you are still mid-attempt.
We build these for specific uses rather than as a general-purpose assistant. Music and language training are the two we work on: the learner speaks or plays, the agent responds to what it actually heard, and the session adjusts from there.
What it can look like
- Conversation practice that adapts to the learner's level
- Spoken pronunciation feedback, given immediately
- Ear training and sight-singing drills that listen back
- Guided practice that keeps a beginner going between lessons
Illustrations of what this covers, not a list of completed projects.
Automation tools
Quietly removing the repetitive, rule-shaped work that eats the day.
Not everything needs a model behind it. A lot of lost time is ordinary repetition: moving information between systems, reformatting it, chasing it, and doing the same checks every week.
Automation here means finding those loops and closing them, using AI only where judgement is genuinely required, and plain deterministic code everywhere else, because it is cheaper and more predictable.
What it can look like
- Moving data between tools that do not talk to each other
- Turning documents and forms into structured records
- Routing incoming requests to the right person automatically
- Scheduled jobs that produce and distribute routine outputs
Illustrations of what this covers, not a list of completed projects.
Data analysis tools
Getting real answers out of the records you already hold.
Most organisations are already sitting on the data needed to answer their own questions. What is missing is a way to ask, without exporting three spreadsheets and rebuilding the same pivot table every month.
We build the layer that sits on top: tools that read your existing records, answer questions in plain language, and produce the summaries and reports you would otherwise assemble by hand.
What it can look like
- Asking questions of your own records in plain language
- Dashboards and recurring reports that build themselves
- Finding patterns and outliers across records too numerous to read
- Turning unstructured notes and documents into analysable data
Illustrations of what this covers, not a list of completed projects.
Custom AI solutions
When nothing off the shelf fits the shape of the problem.
Sometimes the honest answer is that no existing product matches how you work, and bending your process to fit one costs more than it saves.
In that case we design around the constraints that actually apply: your data, your privacy and compliance requirements, the systems you already run, and the people who will use the result every day.
What it can look like
- Internal tools shaped around one team's specific process
- AI features built into a product you already operate
- Systems with strict constraints on where data may go
- Bringing an existing prototype up to something dependable
Illustrations of what this covers, not a list of completed projects.
How a project runs
- 01
Understand the work
A conversation about what actually takes the time, and where the current process bends. Often this is enough to establish whether software is the right answer at all.
- 02
Prototype narrowly
The smallest useful version, aimed at one real task with real data. Narrow scope means you find out quickly whether it works, rather than at the end.
- 03
Put it into real use
Into the hands of the people who will use it, in the systems they already use, with a clear view of what the software is doing and where a person stays in the loop.
- 04
Refine or hand over
Adjust it against how it behaves in practice. You should end up owning something maintainable, not something that only works while we are holding it.