Developer assessment
Give candidates coding problems, multiple choice and written questions in a single assessment. Dextes grades what it can automatically, uses AI to score written answers with its reasoning shown, and hands you a report you can actually defend in a hiring discussion.
No credit card required
Developer hiring has a measurement problem. A CV lists technologies, a portfolio shows finished work, and an interview shows how someone performs when a stranger is watching them think. None of them tells you whether a candidate can read an unfamiliar problem and write something sensible.
The usual answer is a take-home exercise, which is slow to mark and easy to abandon halfway. The other answer is a pure coding test, which measures one narrow thing well and everything else not at all - plenty of strong engineers are better at explaining a trade-off than at solving a timed puzzle.
A useful developer assessment asks more than one kind of question, and gives you something readable at the end.
Four question types, mixed freely inside one assessment.
Candidates solve coding questions in an in-browser Monaco editor, with support for JavaScript, Python, Java, C++, TypeScript, Go and C#. Submissions run against the test cases you define, with optional starter code.
One correct option. Graded automatically the moment the assessment is submitted, with no waiting and nothing to mark.
Several correct options, scored on an exact match so partial guesses do not earn full credit.
Free-text answers for the things multiple choice cannot reach - how someone would design a system, debug a failure, or justify a trade-off.
Psychometric and cognitive assessments are separate assessment types rather than question types - you create them as their own assessment and send them alongside a technical one when the role warrants it.
Three routes, depending on what the question is. The split matters: the parts a machine can settle, it settles.
Automatic
Scored on submission against the correct options. No queue, no marking, no interpretation.
AI-assisted
A model scores the answer against the question and your reference answer, and returns a short explanation of the score. You read the reasoning, not just the number.
Human review
Any AI score can be overridden. Answers the model could not grade are routed to a review queue rather than given a guess.
A graded written answer
Question
Explain how you would debug a service that is slow only under load.
Why this score: the answer identifies connection-pool exhaustion and describes checking metrics before changing anything, but does not mention isolating the slow dependency.
Illustrative example. AI grading is a starting point, not a verdict - the reasoning is shown so a human can disagree with it.
Mix question types, set a time limit, reuse questions from your bank.
Send individually or import a list by CSV. Reminders can go out automatically.
In the browser, with a server-enforced timer and answers saved as they work.
Automatic where it can be, AI-assisted for written answers, human review where needed.
Read the report, shortlist, tag, and export.
Every attempt produces a per-question breakdown: what the candidate answered, what the correct answer was, the points earned, and for written answers the reasoning behind the score. You can see where a candidate lost marks rather than only that they did.
Across a group, the reports show score distribution and per-question performance, which is usually how you find out that a question was badly worded rather than that everyone was weak. Results export to CSV, so a hiring panel can read them without an account.
Candidates can be shortlisted and tagged as you go, so the decision lives next to the evidence for it.
Assessments can carry a time limit that is enforced on the server rather than in the browser, so pausing JavaScript in developer tools does not buy extra time. The countdown a candidate sees is corrected against the server as they work.
Switching away from the assessment tab is detected and recorded on the attempt, and you can set a threshold at which the assessment submits itself. Both are optional and off unless you turn them on.
There is no webcam recording, screen recording or identity verification. If your process requires supervised proctoring, Dextes is not a substitute for it.
Four question types: coding questions solved in an in-browser editor and run against test cases, multiple choice, multi-select, and written answers for reasoning and design questions. A single assessment can mix all four.
JavaScript, Python, Java, C++, TypeScript, Go and C#. Each coding question can carry starter code, and submissions are run against the test cases you define.
An AI model scores each written answer against the question and your reference answer, and returns a short written explanation of why it awarded that score. The explanation appears alongside the score in the report, so you can judge the grading rather than take it on trust.
Yes. Every AI-graded answer can be reviewed and overridden. Answers the model could not grade confidently are routed to a review queue rather than given a guessed score.
Assessments can set a time limit enforced on the server, detect tab switches and record them on the attempt, and optionally submit automatically once a threshold is crossed. There is no webcam or screen recording.
Build an assessment, send it to one candidate, and see what the report looks like before you commit to anything.