Source code and infrastructure
Repositories, configuration and infrastructure definitions. All yours, with no opaque components left in our hands.
Software engineering
AI software development
Most AI projects stop at the demo. The distance between a prototype that works in a meeting and a system that survives daily use is made of integrations, error handling and monitoring — that's the part we do.
A 30-minute technical call, with the people who'd write the code.
- From prototype to production
- On-premise when data can't leave
- Code and documentation handed over
- No platform lock-in
When it helps
When building bespoke is the right call
Buying is almost always faster. Building pays off when one of these applies.
- The process is specific to your company and no product on the market reflects it.
- Data cannot leave your infrastructure, for regulatory or contractual reasons.
- The system has to integrate deeply with existing business systems and databases, not sit beside them.
- A prototype or internal script works, but isn't reliable enough for daily use.
If an off-the-shelf product covers 80% of the need, that's usually the better buy — we'll say so even when it means no project for us.
What we deliver
Software someone else can maintain
A system is finished when it runs without the people who wrote it. That's the standard we hand over against.
Integration with existing systems
Connections to your ERP, databases, internal services and third-party APIs, with error handling and edge cases covered.
Automated quality evaluation
Tests and metrics that tell you whether the system still answers well after a change. Without them, every update is a gamble.
Deployment and monitoring
Containerisation, repeatable releases and visibility into cost, latency and errors once the system is live.
Documentation and handover
Technical documentation and sessions with your team, so maintenance can pass to whoever you choose.
How we work
How a development project runs
Short cycles with something working to look at early, rather than a single delivery at the end.
Technical specification
We define expected behaviour, edge cases, integrations and acceptance criteria. This is the phase that prevents expensive surprises.
First working version
We build the main path end to end, so you can actually use it instead of judging it from a description.
Hardening
Edge cases, performance, running costs and security. This is where a prototype becomes a dependable system.
Production and handover
Go-live, active monitoring and handover. We stay available through the settling-in period.
What we build
The systems we work on
Areas where we've shipped code into production, not just experiments.
AI agents
Systems that carry out multi-step tasks using tools and APIs, with guardrails, spend limits and human oversight at the critical points.
RAG and GraphRAG systems
Retrieval over documents and knowledge graphs, with source citation and measured answer quality.
Machine learning models
Forecasting, classification and anomaly detection: from data preparation to a running model, with retraining planned from the start.
On-premise and edge inference
Models running on your own infrastructure or close to the machine, when latency, cost or confidentiality rule out the cloud.
Computer vision
Image and video analysis for quality control, recognition and counting, including on resource-constrained hardware.
MLOps
Repeatable releases, model versioning, quality monitoring over time and control of inference costs.
Working together
How an engagement is structured
The format depends on how well defined the problem is when we start.
Fixed-scope POC
When technical feasibility is still uncertain: we test the riskiest assumption before committing to full development.
When the question is "can this be done?".
Milestone-based development
The project advances through milestones with verifiable deliveries, so you can assess progress and adjust course along the way.
When the system to build is clear.
Machine learning consulting
When your team develops in-house but needs specific expertise: algorithm selection, evaluation design, code review.
When you need technical depth, not extra hands.
We don't publish rate cards: development cost depends on how many integrations are needed, the state of your data, the reliability requirements and the support period after release. We define the scope on the first call and the proposal arrives with numbers per milestone.
FAQ
The technical questions we hear most
Buy, when a product covers the need: it costs less and starts immediately. Build when the process is specific, the data can't leave, or integration with internal systems is the bulk of the work. It's one of the first things we assess, and sometimes the answer is that no project is needed.
Yes. Open models can run on your infrastructure, on-premise or in a private cloud, and in some cases directly on hardware in the field. It changes the cost profile and achievable performance, so it's a decision to take at the start.
Whichever you prefer. We hand over code and documentation so your team can take it on, and run the handover sessions needed. If you'd rather we maintain it, that's a separate support arrangement.
Both, depending on the case. Commercial models are often the fastest and most capable choice; open models win on confidentiality, high-volume costs and vendor independence. We design systems so the model can be swapped without a rewrite.
With an automated evaluation built alongside the system: a set of representative cases and metrics that tell you, after every change, whether quality went up or down. Without it you can't evolve an AI system safely — and it's the first thing missing in projects we inherit.
Related services
If you need to work out what to build before building it, start here.
AI for business
You don't need to reinvent the company to use AI. We start from a process that costs you hours every week, test it on real data within weeks, and scale only once the numbers justify it.
AI consulting
We help you decide what's worth doing with AI, what isn't, and in what order. Then, if it makes sense, we build it: we're the same people who write the code, so the advice has to survive contact with reality.
AI consultant
Sometimes you don't need to hand over a project — you need someone to think with before deciding. A senior engineer who knows your context and is consistently available, without hiring a full-time role.
Let's talk
Describe the system you have in mind
A technical half hour on feasibility, the integrations required, and the cheapest way to test the riskiest part.