Ubuntu server fleet
InfrastructureProvisioning, hardening, service management, monitoring, patching, and backups. The unglamorous work that decides whether everything above it stays up.
Applied AI & ML
Infrastructure
Behavioral health
We keep the machines running so the product can be about people.
Nexus Research Labs builds and maintains the infrastructure underneath AI products — Ubuntu server fleets, self-hosted language models, web platforms, and open-source software. Right now all of it is pointed at one thing: a behavioral health app that helps people manage anxiety and depression.
System
Person
Two signals finding the same rhythm
Base OS
Ubuntu LTS
Inference
Self-hosted LLMs
Delivery
Web + app services
Licence
Open source, where it counts
current-engagement
Live systems we
deploy and maintain
deploy and maintain
One environment, doing four jobs at once.
Our current engagement is a full-stack build: the hardware layer, the model layer, and the products sitting on top. Everything below runs on infrastructure we provisioned, hardened, and keep an eye on.
Multiple self-hosted language models
AI / MLModels served on our own machines rather than someone else's API — so sensitive conversations stay inside a perimeter we control, and cost per token isn't a quarterly surprise.
Website and application deployment
PlatformBuild pipelines, releases, TLS, DNS, and the routing that puts a product in front of real users without a weekend of downtime.
An open-source social platform
ProductA community layer we host and maintain — because peer support is part of how people actually get better, and because the code being open means nobody has to trust us blindly.
why-behavioral-health
The reason the
infrastructure exists
infrastructure exists
Anxiety and depression are managed daily, not solved once. Software should meet people at that cadence.
The behavioral health app we're building is designed around what a hard day actually looks like: short interactions, no shame in a missed streak, and no interface that punishes you for being unwell. The AI in it is there to help someone notice a pattern — not to play therapist.
That's also why the models are self-hosted. Mental health data is about the most sensitive thing a person can type into a text box, and shipping it to a third-party endpoint was never a trade we were willing to make on a user's behalf.
capabilities
What we take on
for other teams
for other teams
Healthcare is where we want to be. It isn't where we stop.
The same work — servers, models, deployment, maintenance — is the same work whether you're a clinic, a startup, or a team that just inherited a stack nobody documented.
Infra
Deploy and maintain
Linux server provisioning, hardening, monitoring, backups, and the ongoing maintenance that keeps it all boring in the best way.
Models
Run AI on your terms
Self-hosted LLM deployment, inference serving, retrieval pipelines, evaluation, and honest advice about when a model is the wrong tool.
Build
Ship the product
Web applications, APIs, integrations, and release pipelines — built to be handed over, not to keep you dependent on us.
notes
Working notes
from the lab
from the lab
Recent writing
2026 · 07 · 14
Why we self-host every model that touches a user's words
Read the blog
A walk through the threat model we wrote before choosing where inference runs, and what it costs to keep that promise.
2026 · 06 · 28 Designing for someone having a bad dayStreaks, badges, and red notification dots are motivational tools built for people who already feel fine.
get-in-touch