Mock interviews, graded, on your own machine

mockterview runs a realistic technical interview on your own machine and grades it against a rubric it never shows you. Then it tells you exactly where you lost the marks.

Request access See a graded session

The interview

How an interview runs

  • A real conversation. One question at a time, with follow-ups based on what you actually said.
  • Hidden rubric. You cannot answer the marking scheme instead of the question, which is what makes the score mean something.
  • Depth by level. Mid, senior and staff are held to different expectations of the same role.
  • Incident simulation. The system is role-played, symptoms appear only when you go looking, and a complication lands partway through.
  • Three formats. System design, behavioural and incident, each with its own interviewer brief.
  • Long answers. Write as much as you would say out loud, and submit when you are done.
  • Your choice, or its. Pick a role, a level and a format, or let it ask.
Grading

The score, and why you got it

  • Weighted score. Out of 100, built from a score per criterion.
  • Feedback per criterion. Written, and quoting your own answers back to you.
  • A separate grader. Working from the transcript and the rubric rather than an impression of how it went.
  • Specific enough to fix. You get the sentence that cost you the mark, not a note about communication skills.
  • Rubric revealed. After the grade, so you can see what was being looked for all along.
History

Progress over time

  • Every session saved. Transcript and score together.
  • Average per format. And the change from your first attempt to your latest.
  • Reopen anything. Any past interview in full, criterion by criterion.
  • Your file. Back it up, read it with any SQLite tool, or delete it in one command.
Customisation

Roles and rubrics you can edit

  • Roles included. Site reliability engineer, platform engineer and engineering manager, each with focus areas and per-level expectations.
  • Add your own. A new role or format is one file. No code, no rebuild, nothing to wait for.
  • Your hiring bar. Write the rubric your team actually interviews against.
  • Plain text. Roles and rubrics are files you can read, edit and put in version control.
  • Choose the models. A strong one to interview, a cheaper or local one to grade, or the same for both.
  • Set the length. How many questions an interview runs to.
Privacy

Runs on your machine

  • Fully local. Point it at a local model and your answers, your weak spots and your transcripts never leave the machine.
  • Any endpoint. Works with any OpenAI-compatible model, local or hosted, your choice.
  • One binary. Copy a single file onto the machine and run it.
  • No account. Nothing to sign in to and nothing to cancel.
  • No telemetry. Nothing phones home, and nothing stops working when the connection does.
  • Works on a plane. With a local model there is no network step at all.

Request access About the beta