Tech AI Consulting
We research, analyze and build specialized AI systems for teams with a real, defined problem — the kind of rigor behind Anne for Legal, our own product, applied to yours.
We research a problem, analyze what a correct answer actually looks like, and build the specialized system that answers it — then hand over something built to keep running, not to demo well once.
Anne for Legal is the proof, not a case study written after the fact. It’s an AI legal-review workspace we built and run in production, for law firms handling matters that legally cannot be sent to hosted AI. Every observation it makes traces back to a source. Every check it runs has an explicit state — passed, flagged, or unverified, never silently waved through. Nothing becomes final without an attorney’s release. That’s the standard we hold our own work to, and the standard we bring to yours.
What this looks like in practice
Document and data extraction. Invoices, forms, contracts, reports, title records — pulling structured, source-linked data out of unstructured documents so nobody retypes them, and so every extracted value can be traced back to where it came from.
Internal tools. The spreadsheet that outgrew itself and now runs a core part of your operation, with three people editing it and no history of who changed what. Turning it into something with proper access control and an audit trail.
Integration between systems. Two pieces of software that both hold important data and do not talk to each other, so someone maintains both by hand. This is unglamorous work with an immediate payback.
Assistants over your own knowledge. A system that answers questions from your documentation, policies or historical records, with citations back to the source, so answers can be checked rather than trusted blindly.
Workflow automation with a real decision boundary. The sequence of steps that happens every time a matter, a claim, or a client file moves forward — with a clear line between what the system decides and what a human has to.
How we work
Start with the cost, not the technology. The first question is what the current process costs in hours per week, or what it risks when it’s done by hand. If we cannot answer that, we do not have a project yet — we have an idea.
Smallest useful thing first. A narrow version that solves one real case, in production, in weeks. You find out whether it is valuable before committing to the full build.
Built to be maintained. Documented, with the code and infrastructure in accounts you own. If you stop working with us, everything keeps running and someone else can pick it up. Handover should never be a negotiation.
Honest about limits. Language models are unreliable in specific, predictable ways. We design for that — validation, human review where the cost of an error is high, and a clear boundary around what the system decides versus what it recommends. Anyone promising accuracy without that boundary is selling a demo. Anne for Legal is built to that boundary; so is everything we build for clients.
Where AI is the wrong answer
Worth saying plainly, because it comes up often.
If the process is fully deterministic — the same input should always give the same output, with clear rules — conventional software is cheaper, faster and more reliable. If the data is a mess, fix the data first, because a model trained or prompted on inconsistent records produces confident nonsense. If nobody can articulate what a correct answer looks like, no system can be evaluated.
We would rather tell you that in the first conversation than take the project.
Get in touch
Tell us the problem — not the solution you’ve already picked. If it’s a fit, the next step is usually small: a narrow version, in production, in weeks.
Frequently asked questions
- Do we need to know what we want before contacting you?
- It helps to know the problem, not the solution. ‘Our team spends six hours a week copying data between two systems’ is a perfect starting point. ‘We need an AI strategy’ is not, and we will spend the first conversation turning it into something concrete.
- What does a project cost?
- Small automations that remove a recurring manual task typically land in the low thousands. A custom internal tool or a system integration is usually a multi-week engagement. We scope in stages so you can stop after the first one if the value is not there.
- Will this replace our staff?
- In the work we do, almost never. The projects that succeed remove the tedious part of a job — the copying, the retyping, the reconciling — so the people doing it spend their time on the part that needs judgement. Projects that aim to remove people tend to fail for reasons that are more about process than technology.
- Is our data used to train anyone's model?
- Not without you deciding so explicitly. We use commercial API tiers that exclude your data from training by default, and where the data is sensitive we can run models that never leave infrastructure you control — the same architecture behind Anne for Legal. This is a design decision made at the start, not an afterthought.
- What makes this different from a typical AI consultancy?
- We ship a real product, Anne for Legal, under the same discipline we bring to client work: every claim it makes is source-linked, every layer that doesn’t run is reported as unverified rather than silently passed, and nothing becomes final without a human decision. That standard is the pitch, not a slide about it.
Tell us the problem, not the solution
The first conversation is free — it's where we work out whether this is worth doing at all.