LMC Tech
Published Android app
LODORi
A meditation app that writes and voices a personalised session in about 25 seconds. React Native, FastAPI, subscriptions, ads, safety checks, and a public release on Google Play.
Our work / Infrastructure
Self-hosted AI and deploy pipeline
Most of what is on this page runs on hardware we own. Local models keep client data off third-party APIs when that matters; a sandbox lets models write and run code without reaching anything else; and every site deploys from a push to GitHub through a listener that checks, backs up and rolls back.
5
production sites deployed by one signed webhook
12GB
of GPU running Qwen3, Gemma 3, Flux and LTX locally
1
network-isolated sandbox where models may run code, pen-tested
0
commits allowed on production servers: a hook refuses them
The brief
Own the stack, own the risk.
Cloud AI is convenient until a client asks where their data went. A self-hosted box answers that. It also had to be safe to let models loose on, and it had to make deploying boring.
The deploy guards exist because on one August night a fix committed on the server was silently destroyed by the next deploy, and a badly edited environment file took a charity's site down for eight minutes. Both are now impossible by construction.
- Ollama serving Qwen3 and Gemma 3 behind a chat interface, ComfyUI hosting Flux and LTX for image and video, all on one RTX 3060.
- A Docker sandbox with no network, dropped capabilities and per-project valves, plus a broker service so a model can only touch its own folder. Externally pen-tested.
- A Slack channel as a shared bus: agents on three machines post what they deployed, found or decided, so nobody explains the same thing twice.
- A webhook listener that verifies signatures, routes by repository and branch, rescues unpushed work, snapshots PostgreSQL before migrating, runs Django deploy checks and rolls back on failure.
- Scheduled jobs for token health across every social platform, nightly publishing, thermal logging and an incident sentinel.
- Self-hosted, cookie-free analytics for every site we run.
Under the bonnet
Ubuntu, systemd, Docker, nginx, Cloudflare tunnels.
Ubuntu 24.04, systemd user timers, Docker with hardened containers, nginx and Cloudflare tunnels for exposure without open ports, a Python webhook service, Bash deploy scripts with a shared guard library, Umami analytics, and a documented credential store with per-machine read-only mirrors.
- Ubuntu
- Ollama
- ComfyUI
- Docker
- systemd
- nginx
- Cloudflare Tunnel
- GitHub webhooks
- PostgreSQL
- Umami
What it shows
- Private AI for organisations that cannot send data to the cloud.
- Deployment as a safety system, not a ritual.
- Operational maturity that small studios rarely bother with.
AI social media manager
Social Jen
Give it one idea, a product photo or a video and it drafts, schedules and publishes to nine social networks, writes graded Etsy listings and renders your 3D models. Nothing goes out until you approve it.
AI to 3D print pipeline
modelgen
Describe a part or drop in a picture and get a watertight, print-ready STL. Exact parametric CAD written by a language model, or organic meshes from AI, both validated before a slicer ever sees them.
Next step