XenAI Applied AI
Research, put to work.
Applied AI is where our research meets a real setting. We design and build language model systems for organisations that cannot afford a confident wrong answer, and we advise teams building their own.
What we build
Source-bound assistants
Systems that answer from a defined body of sources, show where each claim comes from, and check that the source applied on the date in question.
Risk-controlled pipelines
A small model answers what it can. What it cannot is passed to a stronger check, a larger model or a person. The design goal is an error rate agreed in advance, not discovered afterwards.
Private deployment
Models served on infrastructure the client controls, so that sensitive records stay where they already are.
Where we work
Law
Family law in England and Wales is our first case study. The rules have changed in recent years, so the right answer depends on the date of the events as much as on the source. We build for solicitors, who need both before they can rely on an answer.
Education
Learning tools built with the same care over accuracy and over personal data.
Projects in progress
We are working on these now. None is finished, and we will say so here when one is.
XenLex
Our first case study, in family law. XenLex is a research and document assistant for solicitors in England and Wales. It is being built to check each claim in an answer against published sources and against the date on which they applied, and to hand over to the solicitor what it cannot support.
XenQ
A legal examination system, being built on FastAPI. It is designed to run on a mix of different language models rather than a single one.
XenBundle
A court bundle tool that checks before it files. It is being built to assemble an indexed, paginated bundle and to test it against the rules that apply to that court and that hearing.
To hear more about any of these, write to us.
Write to us