Lunar Labs
We build the thing,
not the slide about it.
A small engineering team shipping AI systems into production — for clients, and for ourselves. CloudHub and Arena run on the same stack, under the same on-call rotation.
Lunar Labs
Software that survives
contact with real users.
A small team that ships whole systems rather than proofs of concept. You get the repository, the infrastructure and the documentation — not a demo that dies when the trial ends.
AI Product Engineering
From a problem statement to a system in production. We build the whole thing: model, data path, interface, and the operational work that keeps it running.
- Next.js and TypeScript applications
- LLM orchestration and evaluation
- Retrieval pipelines over your data
- Postgres, Supabase, row-level security
Agents and Automation
Agents that act inside your systems with an audit trail. Scoped permissions, verifiable steps, and a human where a human belongs.
- Tool-using agents with guardrails
- Document and workflow automation
- Internal copilots on private data
- Evaluation harnesses and regression suites
Computer Vision
The overlap between our two halves. Detection, tracking and scene understanding, trained on footage we can also shoot for you.
- Detection and tracking pipelines
- Video understanding and indexing
- On-device and edge inference
- Synthetic data generation
Platform and Infrastructure
The unglamorous half that decides whether the rest survives contact with real traffic.
- Cloud architecture and IaC
- CI/CD and release engineering
- Observability and cost control
- Security review and hardening
How we build
Four positions we hold.
You own it from commit one
The repository is yours. The infrastructure is in your account. There is no licence back to us and nothing that stops working if you stop paying us.
A demo is not a system
Anything can be made to work once, on stage, with the right input. We build for the second month — error paths, rate limits, migrations, the boring half that decides whether it survives.
Evaluated, not vibed
An AI feature ships with a test set and a score. Otherwise "it got better" is a feeling, and the regression that broke it will go unnoticed for weeks.
We use what we sell
CloudHub runs real events on the same stack we would build yours on. When we recommend something, we have already been on call for it.
Frontend
- Next.js 16
- React 19
- TypeScript
- Tailwind 4
Backend
- Postgres
- Supabase
- Edge functions
- Row Level Security
AI
- Claude
- Retrieval pipelines
- Evaluation harnesses
- Vision models
Platform
- Vercel
- AWS
- Terraform
- GitHub Actions
Products
We do not only build for clients.
We ship our own.
Every product here runs in production with real users on it. They are also the proof: the same team is available to build yours.
Lunar Reel
BetaFootage that indexes itself.
Point it at an archive and get back a searchable library: shots described, faces grouped, moments timestamped. Built from the vision work our own studio needed first.
- Shot-level semantic search
- Automatic selects and rough cuts
- Face and object grouping
- Runs on your storage
How it goes
Four steps, no surprises.
The same process runs both halves of the company, because the thing that goes wrong on a shoot and the thing that goes wrong on a build are usually the same thing: nobody wrote it down.
- 01
One conversation
A call that ends with a date and a number, not a discovery workshop. Whichever half you came for, the same people answer.
- 02
A written plan
Treatment and shot list for a film; a scope, an architecture and a milestone schedule for a build. Both fit on a page you can forward.
- 03
The work
Crew on location, or commits in a repo you own from day one. You see progress weekly, not at the end.
- 04
Delivery that lasts
Masters in every format you need, or a deployed system with the documentation to run it. Then we stay reachable.
Tell us which half you need.
Or both.
A date on the calendar, a camera list, or a system that does not exist yet — it starts the same way.
Al Reem Island, Abu Dhabi — replies within one working day