What we build, and how we scope it.
Four categories of work. Most engagements draw on more than one of them, because an AI feature that nobody can operate is not finished software.
Systems that take action, with a person still in the loop.
Multi-step workflows that read a request, decide what to do, call real tools, and stop for a human when the stakes justify it. The interesting engineering is not the prompt — it is the state machine, the retries, and the audit trail around it.
Capabilities
- Tool-using agents with typed, validated tool schemas
- Multi-step workflow orchestration and durable state
- Human-in-the-loop approval gates and escalation paths
- Internal assistants scoped to a team’s real systems
- Support and operations triage automation
- Trace logging, replay and failure analysis
What you get
- A running workflow in your environment, not a notebook
- An evaluation set that fails loudly when behaviour regresses
- Operator surfaces: queues, approvals, overrides, audit history
- Documented tool contracts and permission boundaries
Answers you can trace back to a source.
Retrieval, document intelligence and conversational interfaces over your own content. We treat retrieval quality as an engineering problem — chunking, ranking, grounding and citation — rather than something the model is expected to sort out on its own.
Capabilities
- Retrieval-augmented generation with citation and grounding checks
- Document intelligence: extraction, classification, structured output
- Enterprise and product search over mixed content
- Conversational analytics over governed data
- LLM feature integration inside existing products
- Retrieval evaluation harnesses and relevance tuning
What you get
- An ingestion pipeline that handles the messy real corpus
- A retrieval layer with measured precision and recall
- Interfaces that show their sources and admit uncertainty
- A confidence policy: when to answer, when to defer
The unglamorous craft that decides whether any of it ships.
We are a software engineering studio first. AI is a capability we reach for when it earns its place — the rest of the time we are designing schemas, building APIs, shaping interfaces and shipping products that work on an ordinary Tuesday.
Capabilities
- SaaS products and MVPs taken from scope to launch
- Web applications and internal tools
- APIs, services and data models
- Dashboards and operational interfaces
- Modernising systems that have outgrown their original design
- Front-end architecture, design systems and accessibility
What you get
- A deployable codebase with tests and a CI pipeline
- Interface design that is implemented, not just specified
- Documentation a future engineer can act on
- A handover that does not depend on us staying
Connecting the systems a business already runs on.
Most useful automation lives in the space between tools that were never designed to talk to each other. We build the connective layer — and the monitoring that tells you when one end of it changes.
Capabilities
- CRM, support desk and billing integrations
- Workflow orchestration across third-party services
- Data pipelines and synchronisation
- Deployment, environments and infrastructure-as-code
- Observability: logging, tracing, alerting, cost monitoring
- Webhook, queue and event-driven architecture
What you get
- Integrations with idempotency and backoff, not best-effort calls
- Secrets handled server-side, with rotation documented
- Dashboards for throughput, error rate and spend
- Runbooks for the failures that will actually happen
Short loops, written decisions.
- 01
Scope, together
A short paid discovery: we learn the actual problem, look at the real data, and write down what we would build and what we would not. You own that document whether or not we continue.
- 02
Build in visible increments
Working software in your hands early and often, with the evaluation set and the trace view in place from the start rather than added once something breaks.
- 03
Hand over properly
Documentation, runbooks and a codebase another engineer can pick up. We are happy to stay on; nothing should depend on us doing so.
Have something difficult in mind?
We like ambitious problems — the ones with awkward constraints, messy data and a real cost attached to getting them wrong. Tell us what you are building and what makes it hard.
- Good fit
- A real workflow with real constraints, an internal owner who can make decisions, and a problem where being wrong actually matters.
- Less good fit
- A chatbot with no defined job, a fixed-scope rebuild of something that already works, or a project that needs to look finished before it is understood.