Five things, and they are the same five whatever the job is.
A website, a mail platform for an organisation that cannot use a foreign provider, or a furnace that has never reported its own temperature. The work underneath is infrastructure, software, and knowing which layer is lying to you.
Managed infrastructure
IaaS and managed hosting on infrastructure we operate: web and application deployment alongside AI/ML workloads, including GPU inference via Modal. Built from the base up rather than from a pre-made image.
Who it is for. Organisations that need somebody to own the machine, and to be able to say which machine it is.
Proxmox · nginxGPU inferenceMonitoring and backups
What you get
Provisioning on our own Proxmox hypervisors
nginx, TLS and DNS configured and documented
Monitoring, backups, and a restore that has been tested
GPU inference via Modal where a workload needs it
Custom software
Go on the backend, Next.js on the front. We write the whole stack, so there is nobody to hand a problem to when something in the middle of it breaks.
Who it is for. Anyone who has been handed a system nobody can explain, or who needs one built that can be.
GoNext.js · TypeScriptPostgres
What you get
A Go backend, and the schema it reads and writes
A Next.js front end, server-rendered
PostgreSQL, with migrations in the repository
Deployment on infrastructure we run, or yours
Fuzon Mail
A self-hosted email platform for organisations that cannot put their correspondence on somebody else’s continent: SMBs, government, and regulated industries.
Who it is for. SMBs, government bodies and regulated industries whose correspondence cannot leave the country.
Postfix · DovecotRspamdData stays in Nepal
What you get
Postfix, Dovecot and Rspamd, tuned rather than boxed
One Fuzon Account across Mail, Drive and Office
Passkey sign-in, so passwords become optional
SPF, DKIM, DMARC and reverse DNS, set properly
IoT and hardware
Sensors, embedded gateways, PLC and control-system integration, and the data pipeline behind them. Instrumenting a process that currently has no numbers coming out of it, and making those numbers trustworthy.
Who it is for. A plant, a process or a facility that currently produces no numbers, or numbers nobody trusts.
Sensors and gatewaysPID and controlTelemetry pipelines
What you get
Sensors and gateways, specified and installed
PID control loops, tuned against the real process
Fault detection, with alerts that mean something
A telemetry pipeline, and the logs behind every reading
AI engineering
Model deployment and inference pipelines, put into production and kept running. Not a proof of concept in a notebook.
Who it is for. Teams with a model that works in a notebook and has to work in production.
ModalInference pipelinesApplied ML
What you get
Inference deployed on Modal GPUs
Provider routing, so one outage is not your outage
Pipelines assembled end to end, not stitched per request
The cost of a run, measured rather than estimated
How we work
Four things that do not change between engagements.
01
We look first
Before a quote, we read what exists: the code, the machines, the process on the floor. A quote written from a brief alone is a quote we will have to revise.
02
One team, whole stack
The same people handle the metal, the backend, the front end and the deployment. Nothing is thrown over a wall, because there is no wall.
03
You get the keys
Repositories, credentials, runbooks and schemas are yours. Nothing we build requires us to keep operating it.
04
We stay on if you want us
Most clients keep us running the infrastructure afterwards. That is a choice, not a lock-in: the work is handed over either way.
We do not publish a price list.
A brochure site, a mail platform for an organisation of two hundred people, and a control system for an induction furnace are not the same engagement, and a page that priced them as if they were would be wrong three times. Describe what you need to run and we will quote it. That includes telling you, where it applies, that you do not need us.