Deliver on time
Missed dates that were never about the work.
4 minutes of waiting becomes 11 seconds
See the evidenceHow I can help
Each of these started as a real business problem at a real team. The numbers are measured, the tools that fix them are open source, and you can check every claim before you speak to me.
Missed dates that were never about the work.
4 minutes of waiting becomes 11 seconds
See the evidenceInfrastructure spend you can explain, and stop paying for idle capacity.
250 MB becomes 50 MB, and 7.9% idle CPU becomes none
See the evidenceCritical processes that keep running when someone else's service does not.
No external dependencies, and every attempt recorded
See the evidenceYou have probably already tried the obvious fix
If any of these sound familiar, the next part is for you.
A bigger, faster runner
The bill goes up, the queue gets shorter, and the same five minutes becomes three. You have moved the cost, not removed it.
A memo telling people to run the checks locally
It works for a fortnight. Then someone pushes from a laptop without the setup, and the pipeline is the only thing that is protected.
Raising the limits, or buying more minutes
It works until the next month. Then the same thing happens again, and this time it is a surprise to everybody.
A commercial mock service
A seat licence, an internet connection, and a runtime the size of a small server. It works — right up until the budget review.
Deliver on time
A business needs its delivery dates to mean something. Right now they are blown for a reason nobody can name, and the reason is not the work — it is the queue, and sometimes it is the bill.
What is happening
A developer makes a change, pushes it, and waits four minutes for the build. Twenty developers, twenty changes a day, two hundred and fifty working days a year. That is over sixteen hundred hours a year spent looking at a progress bar, and it is a salary-sized cost that appears on no invoice. It gets worse when the shared build machine runs out of budget: then the wait is not four minutes, it is forever, and the pipeline looks broken rather than exhausted.
What I do
The check now runs on the developer's own laptop, before the code is sent anywhere. The result comes with it, so nothing is left unverified, and the answer never depends on a shared machine being available.
CI Kit runs the same scripts the CI workflow runs, on the machine in front of the developer, with the toolchain pinned so the two cannot disagree. It is a shell tool with no runtime to install.
The wait drops from four minutes to eleven seconds, and the result arrives every time. Dates stop being missed for reasons that have nothing to do with the work.
At a loaded cost of fifty pounds an hour, sixteen hundred hours of waiting is roughly eighty thousand pounds a year, for a tool that costs nothing to run and nothing to license. The waiting is only part of it — the same limit stops the pipeline running at all, which is a far more expensive failure.
4m 12s becomes 11s. Job retries fall from 8.4% to 0.6%. Twenty-eight builds that had been unverified for a week get verified again — and the checks that move to the laptop stop consuming metered build minutes altogether.
Control the cost
A business needs to know that the part of the product which takes money is the part that was tested. Usually it is the other way round, and nobody planned it that way.
What is happening
To test a payment, the team needs a live account with a payment provider, a working network, and somebody's patience. So the payment code gets tested late, by one person, manually — and the manual instructions live in that person's head. The rest of the team cannot test it at all, so they do not, so it breaks.
What I do
A stand-in for the outside service runs inside the test, on your own machine, with no account and no internet. It behaves like the real thing, including the ways the real thing fails — the timeouts, the declined cards, the webhook that arrives twice.
api-mock-services is one dependency-free Node process standing in for eighteen providers, with the memory footprint enforced as a build gate rather than promised in a README.
Payment and identity paths get tested on every single change, and the test environment fits in fifty megabytes instead of four containers and a quarter of a gigabyte.
Four containers at 250 MB, idling at 7.9% of a core on every laptop, is a quarter gigabyte of every developer's machine multiplied by every developer. One process at 50 MB, with no idle CPU and no dependencies to install, renew, or audit.
4 processes become 1. 250 MB becomes 50 MB. 7.9% idle CPU becomes 0.0%. The first request is answered in 1.4 seconds instead of 20.
Remove the dependency risk
A business needs its critical processes to keep running, and to be able to prove what happened afterwards. That is a hard requirement in banking, healthcare, and government — and a good idea everywhere else.
What is happening
On a secure or isolated network, anything that reaches out to a cloud service is a single point of failure. The usual response is a workaround nobody documents, and the audit trail is reconstructed from logs after the fact, if at all.
What I do
A background service that runs your workflows with no internet at all. The steps are ordinary data files, so a manager can read them, and every attempt is written down for whoever checks later.
The Automation Gateway is a Windows Service that executes JSON blueprints inside a runspace pool, with plug-in auto-discovery, role-based permissions, break-glass toggles and a transactional per-attempt audit trail.
The work carries on regardless of the network, and a step that failed and healed itself is still visible in the record — which is the difference between an incident and an argument.
An outage that is somebody else's problem becomes an outage you can see, trace and explain to an auditor. That is the difference between a day of explaining and a day of proving.
No external dependencies and no build step. Every attempt, including the retries, is written to a transactional database.
How I work
Four habits, applied to every project.
Everything is published under a permissive licence, with the code, the tests and the measurements in the same repository. If it is not good enough to publish, it is not finished — and you can read it before you commit to anything.
Each project has an evidence document saying how every figure was taken, publishing the measurement that was thrown away, and listing the claims that are not made. You do not have to take my word for it.
Small, quiet, and no licence to renew. The mock engine takes 50 MB and no idle CPU, against 250 MB and 7.9%. The local gate costs nothing, because it does not need a machine you are paying for.
Documented, tested, released and versioned. If I disappear tomorrow, the tooling keeps running and somebody else can pick it up. That is the point of open source, and it is also the point of a supplier who is not a single point of failure.
Written for engineers, but the first paragraph is for everyone else. Every tool mentioned here is open source, so you can read the code, run the tests and check the numbers before you decide anything.