Own the perimeter
PrincipleSensitive data stays on infrastructure the client controls. If a convenience means sending user data somewhere we can't see, it isn't a convenience — it's a liability with better marketing.
about
A small lab that runs what it builds.
Nexus Research Labs is an applied AI and infrastructure practice. We deploy and maintain the systems our software runs on, which means we live with every decision we make — and it keeps us honest about the ones we recommend to you.
what-we're-building
The long project
The work that started this
Our current engagement is a behavioral health application for people managing anxiety and depression. Underneath it sits a fleet of Ubuntu servers we provisioned and maintain, several self-hosted language models, the deployment pipeline that ships the product, and an open-source social platform for peer support.
We took on the whole stack deliberately. When one team owns the servers, the models, and the product, there's nowhere for a privacy decision to get quietly lost in a handoff.
That project shapes how we work everywhere else. Building for someone in the middle of a panic attack teaches you things about latency, clarity, and restraint that a dashboard project never will.
We intend to keep working in healthcare. We also take on AI and infrastructure work in any sector — the discipline transfers cleanly.
principles
How we make
the calls
the calls
What we hold to
Boring where it matters
PrincipleNovel architecture is fun. Stable, well-understood, well-documented systems are what let a small team run production without burning out. We save the ambition for the product layer.
Build to hand over
PrincipleEvery environment we set up comes with documentation clear enough for someone else to take it over. A client who can leave and chooses to stay is a better client relationship than a locked door.
Say when AI is the wrong answer
PrincipleWe run language models daily and we know their edges. Sometimes the honest recommendation is a database query, a form, or a person. You'll get that recommendation.
stack
What we reach
for first
for first
Working stack
Systems
- Ubuntu LTS
- systemd, nginx
- Containerised services
- Automated backups
Models
- Self-hosted open-weight LLMs
- Local inference serving
- Vector search and retrieval
- Evaluation harnesses
Product
- Modern web front-ends
- APIs and background workers
- Relational data stores
- Open-source platforms
Tools are means, not identity. If your team already runs something well, we'd rather learn it than replace it.
contact
Start here
Tell us what's running — or what should be.
A short description of your systems and what you're trying to reach is enough to start. If we're not the right fit, we'll say so in the first reply.
What to include
- What you're running today
- What's breaking or missing
- Any compliance or privacy constraints
- Rough timeline
get-in-touch