// Cloud and scale
Infrastructuresized for real traffic
Cloud setup, pipelines and performance work, including the unglamorous job of making something that is already live stop falling over.
Timeline
2 to 8 weeks
You get
Infrastructure, pipelines, runbook
Scope
Fixed after an audit
Ownership
Your cloud accounts
Two jobs live here. Setting up infrastructure for something being built, and going into something already running that is slow, expensive or unreliable, and finding out why.
The second one starts with measurement every time. The component everyone blames is usually not the one costing the time, and rewriting it is the most expensive way to learn that.
What the work covers
Cloud setup and deployment pipelines
Environments, CI and release configured so deploying is boring. Rollback that has been tested rather than assumed.
Performance and Core Web Vitals
Field data first, then the fix. Measured before and after, on the devices and connections your users actually have.
Architecture for load
Caching, queues and read paths designed for the traffic you have and the traffic you expect, not for a diagram.
Database design and optimisation
Schema, indexes and query paths. Most systems described as needing more servers need one index.
Cost control
Cloud spend traced to the thing spending it. The largest line on a cloud bill is very often something nobody meant to switch on.
Rescue and modernisation
Systems already in production, changed in place. Strangled a piece at a time so the business keeps running throughout.
What goes wrong with infrastructure
Not a list of things other people do badly. These are the ones worth checking before anybody quotes you.
- 01Optimising the part that felt slow. Without a profile, the fix lands on the wrong component and the system is now more complicated and just as slow.
- 02A rollback path nobody has tested. It is discovered to be broken during the incident it exists for.
- 03Autoscaling used to hide an unindexed query. It works, it is billed monthly, and it never gets found.
- 04Lab scores accepted as performance. Lighthouse on a fast machine says one thing, field data from real phones says another, and Google uses the second.
- 05A rewrite chosen when a migration would do. The rewrite takes a year, the old system still has to be maintained, and both are half-supported for the duration.
Questions we get asked
With measurement, not with code. Field data from real users, then a profile of the server path, and only then a fix. The component everyone blames is usually not the one costing the time.
Often paired with
Your next product starts here.
Tell us what you're building, right here. Attach a brief if you have one — no commitment, no pressure, just a real reply.