Machine learning and AI, built into the way you work.
We build models and AI features into your products and workflows, such as document checks, classification, search and assistants that work with your own data, with review steps, logging and access controls, so you can always see what the AI did.
What we build
AI where it clearly helps, and a simpler approach when that would do better.
-
Classification and prediction
Models that sort, score and predict, such as routing requests, flagging risk or forecasting demand.
-
Matching and recommendations
Matching records, people or products, and recommending what is most relevant.
-
Document and identity checks
Reading documents, pulling out the details that matter and checking identity, like the checks in our IDClear product.
-
Search across your own content
Search that understands questions, across your documents, knowledge base and records.
-
AI assistants connected to your tools
Assistants that answer questions and complete tasks using your own data and systems, securely and within limits you set.
-
Automation with a person in the loop
AI that prepares the work, such as drafting, sorting or data entry, while people approve the result.
-
Deployment and monitoring
Models put into production with monitoring, so they keep working and are updated when the data changes.
What goes into every AI feature we build
Behind this work
IDClear, our identity verification product, runs document and identity checks. Johnson, one of our co-founders, is doing a PhD at the University of South Africa on AI and software delivery in the public sector.
-
Human review
Review steps where decisions matter, built in from the start. It is also one of the topics in our co-founder's PhD research.
-
Transparent
Logs of what the AI saw, suggested and did, so results can be explained and checked.
-
Access controls
The AI sees only the data it needs, with the same permissions as the person using it.
-
Measured
Accuracy tested against real examples before launch, and watched afterwards.
-
Private
We agree where your data is processed and stored, and keep it there.
-
Practical
We use AI where it clearly helps, and say so when a simpler approach would do better.
Technology we use
We'll suggest the simplest set of tools that does the job, and that your own team can live with after we're done.
Languages
- Python
Approaches
- Machine learning models
- Language models
- Search over your own documents
Cloud
- AWS
- Azure
- Google Cloud
What you get
Everything your team needs to run it, change it and understand it.
- Models or AI features running in production
- An evaluation of accuracy against real examples
- Review steps, logging and access controls
- Monitoring of accuracy after launch
- Documentation of how the model works and where its limits are
- A plan for keeping the model current
How a project runs
Nothing starts without your sign-off, and you'll see working software every two weeks.
- 1
A free call
Thirty minutes to talk through what you need. If we're not the right fit, we'll say so.
- 2
Discovery
A few short workshops to agree goals, users, constraints and what success looks like.
- 3
Design
Screens you can click through and test with real users before we write any code.
- 4
Build
Two-week sprints. Each one ends with a demo of working software you can try and comment on.
- 5
Launch
Automated tests, security checks and a go-live plan with a way back if something goes wrong.
- 6
Looking after it
Your system moves onto a Managed Services retainer, supported by the people who built it.
What happens after launch?
You can keep us on to look after it. The people who wrote the code are the ones who keep it running, so nothing gets lost in a handover.
Common questions
- How do you keep AI accountable?
- Every feature has review steps where decisions matter, logs of what the AI did, and access limited to what it needs.
- Can AI work with our confidential data?
- Yes, with care. We agree where data is processed, limit what the AI can see and log its use. Some projects run entirely inside your own cloud account.
- How do you charge?
- Build work is billed by the hour against an estimate we agree before we start, and you see hours and progress every sprint. Looking after a live system is a monthly or yearly retainer.
- Where do we start?
- With a free 30-minute call. If we're not the right fit, we'll tell you. If we are, we agree goals and an estimate before any work begins.
Tell us what you're working on
The first call is free and there's no sales pitch. We reply within 24 hours.