Truemend started with an audit.
Earlier this year I was asked to review a codebase that had been written mostly with AI tools. I went through it by hand, function by function. It took a long time, and most of what I found wasn’t hard to fix. It was the same problems over and over:
- Functions that ran far too long and did several jobs at once
- Error handling copied from one place to the next instead of written once
- Types set to any so the code would compile, which meant the type checker couldn’t catch anything
- React components with so many useState and useEffect calls that nobody could follow the state
- Debug logging and dead code left behind after the feature shipped
The tools I had could point at some of this. None of them fixed it. I’d find an issue, write it up, and someone still had to go do the work.
I wanted to remove that step.
So I wrote the specification for a tool that would do that work. I called it Smelt. The rule was simple. If a fix has exactly one right answer, apply it. If it can’t, show it to a person and let them decide. It doesn’t guess.
Smelt was taken. I went through a lot of names after that, and my favourite was taken too. Truemend was available. To true something is a machinist’s term for bringing it back to straight. Mend is the fix.
Early on it was clear that companies in regulated industries won’t send their source code to someone else’s servers. So for core analysis, your code never leaves your machine.
Along the way Truemend also became the name for my training and engineering work. The goal is the same in all three: teams fix their own code and don’t need outside help to keep doing it.
Al SajooAI Engineer
From one audit to one name.
- 2026The auditA codebase written mostly with AI tools, read by hand, function by function.
- The ruleA specification for Smelt: apply what can be verified, show the rest to a person.
- The nameSmelt was taken. Truemend was available.
- Local firstRegulated teams won’t send source to someone else’s servers, so core analysis stays on your machine.
- Jul 2026Measured in publicRules that fail blind review are switched off, with the reason published.
- Sep 2026It scores itselfTruemend’s own source: 4,490 findings at its baseline, 1,657 when re-measured.
- TodayOne nameThe product, the consulting and the training work under one name, to one goal.
Two words, one job.
An engineer who reads the code first.
I’m an AI engineer. Most of my recent work has been inside codebases written quickly with AI assistance: a TypeScript service assembled from three acquired codebases, and a learning system built across seven languages on the Microsoft stack.
In both, the code worked. The review queue was the bottleneck, senior engineers read every merge, and nobody could say with confidence where the risk was. Truemend is built around that pattern.
I still do the reading myself. When I take on a codebase, I run Truemend on it, go through what it reports, and sit down with the people who own the code.