Zen Tests is our own product for HR and education. An author describes a topic, dictates it or uploads a file. Zen Tests drafts a test, quiz or form with questions, answer options and a scoring scale. AI grades the answers and writes feedback for each candidate. A score that decides a job or a certificate has to hold up, so every answer records how it was graded: against the answer key, by AI with its explanation, or by a person. The candidate can contest a score, and the test’s author reviews it. AI credits are deducted before the work runs, and the spend is recorded per test and per user
When a score affects someone’s job or certificate, it has to show how it was reached. The candidate has to be able to dispute it
Zen Tests is for people who write tests and check them: HR teams assessing candidates, teachers and training providers testing students, coaches. Without it, the author writes the questions, the answer options and the scoring scale by hand. Then the author reads every answer and writes the conclusions.
The quick fix is to paste the answers into an AI chat. It saves time. It cannot answer the questions that come next. Where did the candidate’s answers go? Which model gave this score, and why? What do we show a candidate who disputes it? Who agreed to which version of the privacy policy? A data-protection officer asks these questions. When a score decides a job or a certificate, the candidate may ask them too.
We built Zen Tests so each of those questions has an answer stored with the result. The consent log, the record of how each answer was graded and the dispute process were built in the first months, before the features that rely on them.
The author describes the topic, the audience and the number of questions. Instead of describing, the author can upload a PDF, Word or Excel file, or dictate. Four AI assistants work in order: a brief, then the questions and answer options, then the scoring scale, then publishing. The author can edit each step by hand or ask the AI to redo it. Near-duplicate questions are removed automatically. The conversation is saved on the test, so the author can stop and come back later. Each step of generation is recorded, so when a test comes out wrong we can see which step failed.
The candidate opens a link or an email invitation. Answers by keyboard or by voice, with a timer if the test has one. After submitting, the candidate gets a score, a grade, a breakdown by section, written feedback, a PDF and an email. The candidate can ask follow-up questions about the result, and can contest it.
Choice answers are checked against the answer key. Text answers are graded by AI when the author turns it on, and by a person otherwise. A person can always grade by hand. Every answer records which of the three it was, and the AI’s explanation when AI graded it. The test is saved with each submission as it stood that day, so a later edit does not change an old result.
A small separate model screens every text answer before the AI grades it. An answer that tries to steer the grader instead of answering is marked incorrect, with the reason. A model does this rather than a list of banned phrases, so the check works in all three languages and catches a rephrased instruction.
A candidate who disagrees contests the score. The test’s author reviews it, keeps or changes the score, and the outcome is recorded and emailed to the candidate. Each consent is written to a log with the policy version and the time. Rows are only ever added: a withdrawal adds a row and erases nothing. Users export their data or delete their account themselves, with 30 days to undo. When a question touches health or other sensitive personal data, the author sees a warning while writing it. The warning does not block.
Every AI action costs credits. Credits are deducted before the work runs. The spend is recorded per test and per user. Cyrillic text takes about twice as many AI tokens per character, so it costs twice the credits per character.
What the product does today. We do not publish usage figures
Vitalii is one of the best front-end freelancers that I have worked with. He is always at the top of my list when I need some help with projects. Very knowledgeable, friendly and easy to work with. Very happy, as always.
Honestly the best dev that I use here at Upwork. He is without a doubt my favorite and preferred programmer. His work is clean, logical and on time. Beyond that he usually identifies missing needs and features in projects that I had not considered, which is massively helpful.
Absolutely impressed with his abilities and work ethic. As a programmer myself, I needed to outsource some workload for a personal side project. I found that Vitalii was knowledgeable and very good at what he does. Also he speaks/writes fluent English, which was super helpful. I certainly will be using Vitalii again on future projects. He has earned my trust. He was actually the 2nd person I hired for this job. The first one could not complete this task as needed (even though they had a compelling resume).
Anyway, I can't say enough good things about my experience with Vitalii. If you are looking for someone who can actually do what they say they can, is easy to work with and completes projects on time, I recommend him fully.
Wow, wow, wow. Vitalii was the best freelancer I have used. So dedicated, so detailed and never, never gave me a hard time even when he was right to do so. A real friend and superb coder, professional. I can't recommend him highly enough!
He is always great to work with and always goes the extra mile. Always happy with his work and communication.
Amazing work as always!
A 30-minute call about one process in your business. We will say where automation fits and where it does not