We find where your AI breaks before your users do.
Elenchi trains people across Africa to stress-test, evaluate and red-team AI products — and places that talent with the teams building them.
The assistant said it checked a link. It never did.
Everyone is shipping AI. Almost no one is testing it well.
Assistants and agents reach production faster than anyone can check them. The people who can reliably break them — and explain the failure to an engineer — are scarce and expensive. Meanwhile a large, capable talent pool sits underused.
Agents ship untested
Hallucinated tool calls, looping replies, lost context, over-cautious refusals — the failures that reach real users rarely have an owner inside the team.
Evaluation talent is rare
Finding a failure is easy to claim and hard to do. Teams struggle to hire people who can red-team a system and write a report engineers can act on.
The talent exists
Africa has a deep, trainable pool of sharp, detail-driven people. Trained to the right standard, they do this work as well as anyone — at a cost that scales.
Train the talent. Connect it to teams. Build on top of it.
One pipeline. We start where we have an edge — AI testing and evaluation — and expand from there.
We train
We prepare people across Africa to test, evaluate and red-team AI systems — not in theory, but on real products, judged by a real teardown you can inspect.
We connect
We place vetted evaluators and AI-quality specialists with teams building assistants and agents — sourced from Africa, working with companies worldwide.
We build
Over time we turn what this talent learns into AI-native products of our own — starting with the quality and evaluation tools we wish every team already had.
- 6+ years in QA & test automation across mobile, web and APIs.
- First QA hire at an AI companion app, testing it in production.
- Built QA frameworks from scratch at several companies. ISTQB-certified.
- Stack: TypeScript, Java, Kotlin, Playwright, Appium, REST Assured.
I find these failures myself. Then I train others to the same standard.
Testing an AI assistant in production, I kept finding the failures teams miss: it claimed to check a link it never opened, sent near-duplicate replies in a loop, lost the thread of a conversation, and flagged ordinary requests as unsafe.
Most teams have no one who owns finding these. I do — and I've built and led teams before, so I can teach it.
Elenchus: the Socratic method of testing a claim by questioning it until the contradiction shows. That is exactly the work.
Not a résumé marketplace. A standard you can inspect.
Proof before you commit
You see a real teardown of a real system before anyone joins your team. Judge the work, not a CV.
Managed quality, not just bodies
We stand behind the output. The product is the quality of the evaluation, not the headcount.
Underused talent, global teams
Sourced and trained from Africa, delivering for companies worldwide. The cost advantage is real; the standard is the point.
Have an AI product? Let us try to break it.
Send us something with an assistant or agent in it. We'll find a real, reproducible failure and show you the report — free for the first teardown.
victor@elenchiai.com