Mock Interviews
Published Jul 4, 2026
Mock interviews in Praxis924 let you rehearse a full technical loop against an AI interviewer that asks a question, listens to your answer, and adapts its next question to how you responded — then closes the session with a /100 score broken down like a real debrief. It's for any learner who wants reps before the real thing: to practice thinking out loud, to see where your explanations get thin, and to get an honest, structured read on where you stand. This walkthrough takes you from opening the feature to reading your scorecard and turning it into a study plan.
What you'll need
- A Praxis924 account you're logged into (register and verify your email if you haven't).
- A quiet 15–30 minutes — the loop is meant to run start to finish in one sitting.
- Ideally, a lesson or two already read in
/learnon the topic you want to be interviewed on, so you have material to draw from.
That's it. There's no setup, no scheduling, and nothing to install. The interviewer runs on the same LLMService that powers the rest of the platform, so as long as generation is working elsewhere, mock interviews work.
How the loop actually works
Under the hood, the mock interview is stateless. Each turn, the app sends the whole transcript so far to POST /interview/turn, and the model reads the full conversation before deciding what to ask next. That's what makes the follow-ups feel real: if your answer to a system-design prompt glossed over failure handling, the next question probes exactly that gap. Nothing is stored mid-session on the server — the client replays the transcript — so the interview is driven entirely by what you've said.
After a set number of answers (N), the model switches from asking to grading. It returns a single /100 score plus sub-scores so you can see why you landed where you did, not just the number.
Step 1 — Start a session
From the dashboard, open the Mock Interview area. Pick the topic or role you want to be interviewed on. Give a real answer to the first question — full sentences, out loud in your head or typed as you would speak. The quality of the adaptive follow-ups depends on the substance of your answers: one-word replies give the interviewer nothing to dig into.
Step 2 — Work through the adaptive follow-ups
Answer each question, then submit. Watch how the next question responds to what you just said:
- Go deep on a concept and the interviewer will push you toward edge cases or trade-offs.
- Stay surface-level and it will ask you to justify or elaborate.
- Mention a specific technology and it may ask you to compare it to an alternative.
Treat it like a conversation, not a quiz. If you'd say "it depends" in a real interview, say why it depends here.
Tip: Narrate your reasoning, not just your conclusion. The communication and depth sub-scores reward the how and why, and the interviewer can only follow up on reasoning you actually make visible.
Step 3 — Read your /100 debrief
Once you've answered N questions, the model scores the whole transcript. Instead of a bare number, you get a breakdown by signal so you know exactly what to fix:
| Sub-score | What it measures | How to move it |
|---|---|---|
| Correctness | Were your answers technically right? | Re-read the relevant /learn lesson; verify claims |
| Depth | Did you go beyond the surface? | Cover trade-offs, edge cases, and failure modes |
| Communication | Was your reasoning clear and structured? | Lead with the answer, then justify it step by step |
| Time management | Did you pace the loop well? | Don't over-invest in one question; keep momentum |
Read the breakdown before you react to the total. A 72 with strong correctness but weak communication is a very different homework assignment than a 72 with the reverse.
When it goes wrong
Because the interview replays the transcript every turn, most problems are about the loop or the provider, not your answers.
| Symptom | Likely cause | What to do |
|---|---|---|
| Follow-ups feel generic | Answers were too short to adapt to | Give fuller answers; the model follows up on substance |
| Turn fails or stalls | LLM provider quota/outage | Wait a moment and resubmit; failover usually recovers it |
| Score seems harsh | Depth/communication under-weighted by you | Read the sub-scores — they show the specific gap |
| Session "lost" on refresh | Transcript is client-held (stateless) | Finish in one sitting; don't reload mid-loop |
Note: The interview holds no server-side session state, so a page reload or navigating away can drop the in-progress transcript. Run each interview end to end without leaving the screen, and let the final score persist before you close it.
What you get
A completed mock interview gives you three concrete things: a realistic rep at thinking under questioning, a /100 score you can track over time, and a sub-score breakdown that turns a vague "I need to study more" into a specific list — shore up correctness on the topics you got wrong, or drill communication by rehearsing how you open an answer.
Next
Feed the debrief back into the platform. Send the weak areas to Lucy, the AI teacher, by opening the relevant lesson and asking her to drill you on exactly what the interviewer flagged — she'll ground her answers in that lesson's own material. Re-read the sections you fumbled in /learn, practice the mechanics in /practice/[lessonId], and let your progress streak and spaced-review schedule resurface the topic. Then run the loop again in a few days and watch the sub-scores move.