// nuwright
london · est. 2026

Expert judgment, shared.

There are never enough experts: clinicians to read every message, astronomers to check every signal.

Nuwright builds small AI models that learn how those experts decide, and wraps them in rules a machine can check. The model flags, the code decides, the expert sets the bar. It all runs on your own computers.

We publish our research and release code openly where we can.

Now building

sen · crisis classifier · BuildingSpots a crisis in milliseconds. The data never leaves your building.Sen is a small model that reads a young person’s message and flags it the way a crisis clinician would, in the time it takes to send. It runs on your own server, so nothing leaves your organisation.
<1 GBtarget size
<150 mstarget answer, CPU
Localruns where the data lives · nothing leaves

What we do

all work →
01Crisis teams receive more messages than people can read at once. sen reads each one on the service's own computer and flags the ones a clinician would want to see first, so nothing is sent anywhere else.care · judgment model◔ buildingnot yet
02Children often find it easier to tap a mood than to say how they feel. The Lift collects short wellbeing check-ins from children and turns them into a clear summary for the practitioners and parents around them, with ideas for the next conversation.paediatric wellbeing · platform● livethe-lift.org ↗
03Organizations that run support chatbots need them to sound like their own staff and to stay safe. relationalAI helps set the bot's tone, reviews its conversations for problems and packages the approved ones so the organization can improve its bot.behavioural health · platform● livepe-rai.vercel.app ↗
04People who contact a non-profit often need help quickly and do not know where to find it. The navigator reads their message, in English or Spanish, and points them to a real local service, or says plainly when nothing fits.non-profit · AI tooling◐ piloton request
05Near-peer advisors help students through the college process, and their best answers are worth writing down. We turned advisors' know-how into a scoring guide and 500 example answers that an advising chatbot can be built from.education · advising● deliverednot yet
1–5 of 11 projects

Funders and partners

Partners are described by domain on every page; only funders are named. The portraits above are examples of thought leaders, not partners.

Much of our work is with medical schools, hospitals and NGOs and can’t be shown yet: it supports ethics applications and papers still in progress.

Work with us

We’re recruiting academics and ML engineers who want to turn expert judgment into small, measurable systems. Partners and funders welcome too.

as · ml engineer · no tracking, no list
Or write directly: hello@nuwright.com