Setting up your interview room

AI interview coachLive voice, video & code

Master everyinterview.

Scored practice in 5 formats and 8 languages. Premium adds a live AI interviewer, on-device presence coaching, and code graded by running it.

  • Live voice
  • On-device video
  • Graded code
Start freeSee features

Free to start. No card needed.

5interview types
8languages
  • Video never leaves your device
  • We never store voice recordings, only the transcript
Sample session
Why this question
Targets your weakest skill: Structure
Live interview
00:00
  1. Q1
  2. Follow-up
  3. Wrap-up
AI interviewerListening

You · candidateUp next: your answer

Sample interview transcript. Interviewer: Hi, I'm your interviewer. This will take about a minute. Interviewer: Tell me about a time you led a team through a hard deadline. Candidate: Last spring we had two weeks to ship a payments launch. I split the work into three owned streams, ran a short daily check-in, and cut two nice-to-haves early. We shipped on time, with zero rollbacks. Interviewer: What would you do differently next time? Candidate: I'd flag the scope risk in week one instead of week two, so the trade-offs aren't rushed. Interviewer: That concludes our interview. Thanks — your scores are ready. Sample scores: Structure 21 out of 25, Examples 19 out of 25, Communication 22 out of 25. Strong answer · ready for next round.
Delivery
–wpm pace
–filler words
–confidence
Sample session
Premium · 4 capabilities

Built different, not just priced higher.

These four aren’t feature-flag upsells. Each one is a different architecture — native audio instead of a relay, on-device inference instead of an upload, independent scorers instead of one opinion, real execution instead of an AI’s guess.

Live voice · Premium

A real spoken interview, not a script reader

Native audio-to-audio with Gemini Live — the interviewer hears you and responds in real time, not a text reply stitched into a text-to-speech relay. Your voice goes straight from your browser to the model over an encrypted connection; our server never receives the audio.

Why it matters

Most voice-practice tools are a cascaded pipeline: speech-to-text → an LLM → text-to-speech, bolted together after the fact. That adds a beat of lag and a flatter voice every hop. Native audio-to-audio skips all three translations.

On-device presence · Premium

Presence coaching that never leaves your device

Eye contact, posture and pace are measured by a face-landmark model running locally in your browser tab — objective signals only, never an emotional read of you. Only the resulting numbers — a percentage, a stability score — are ever sent to our server. The video itself is never uploaded, never stored, never seen by a human.

Why it matters

The common pattern for camera-based coaching tools is: record, upload, wait for a cloud model to process it. Running the model on-device instead removes the upload step entirely — there is no clip sitting on a server to leak, subpoena, or get breached.

Multi-persona panel · Premium

A hiring committee, not one AI's opinion

3-4 interviewer archetypes — recruiter, hiring manager, bar raiser, tech lead — alternate questions, each scoring you on their own rubric. The results converge into a verdict that names both the strongest advocate and the toughest dissenter, the way a real debrief actually works.

Why it matters

A single AI score is one opinion wearing a lab coat. A panel with independent rubrics and visible disagreement is closer to what an actual interview loop feels like — and it is honest about where the panel disagreed, not just the average.

Proctored coding · Premium

Graded by running your code, not by an AI reading it

Your solution executes for real, in a sandbox in your own browser, against test cases derived from a canonical solution before you ever see the problem. Correctness is a measured pass rate — never an AI's guess at whether your code "looks right."

Why it matters

Asking an LLM "does this solution look correct?" is exactly the sycophancy problem this product is built to avoid. Execution either passes or it does not — there is nothing to guess and nothing to flatter.

The one rule every score follows

Every score this product gives you is either measured or withheld — it never guesses, and it never flatters.

If there isn’t enough signal to grade something honestly — too little webcam coverage, too short a clip — that score comes back null, not a generous guess. See exactly how in Data you can trust.

The engine · 6 real algorithms

Not a chatbot. An engine.

Most interview tools are a single AI prompt in a box. We run a layer of real, deterministic algorithms on top — the kind behind chess ratings, spaced-repetition apps and recruiter screening systems. Explainable, instant, and computed from your data.

Elo rating systemthe system behind chess & ranked ladders

Adaptive difficulty

Every answer updates a hidden Elo rating and re-targets the next question to your exact level — no wasted easy reps, no demoralizing wall.

SM-2-derivedthe ease-factor update from the spaced-repetition algorithm Anki is built on

Spaced repetition

Tracks which interview type is fading from memory using SM-2’s ease-factor update, then resurfaces it for a re-drill right before you would have forgotten it. Our interval rule is our own — not canonical SM-2.

Keyword-overlap coveragehow recruiter ATS software parses resumes

Resume match

A deterministic coverage score for any target role — the exact matched and missing keywords, computed locally with zero AI guesswork.

Least-squares regressionthe same trendline math used in forecasting

Readiness forecast

A least-squares regression line through your past scores projects where you will land by your interview date, so you always know if you are on track.

Hamilton apportionmentthe largest-remainder method used to allocate parliamentary seats

Panel question scheduling

Multi-persona Panel mode divides your questions across interviewers by quota, weighted toward whichever persona owns your weakest skill — the same fair-apportionment math, repurposed for interview scheduling.

ε-greedy banditthe explore/exploit rule behind multi-armed-bandit optimization

Coding-topic selection

The Coding Assessment drills your weakest topic 90% of the time and spends the other 10% sampling a fresh one — so practice stays targeted without ever going stale.

Custom built

Our own algorithms — not borrowed from any textbook

Everything above is a famous method we point at honestly. These four we built ourselves, specifically for this product — here is what each one does and where you will see it.

Panel Verdict

Multi-persona panel → final verdict

Turns a full panel interview into one clear hiring decision, and names which interviewer backed you hardest and which one pushed back.

Readiness Score

Goal engine → readiness projection

The single number that answers whether you are actually on track for your target role and interview date.

Presence Score

Live video mock → presence coaching

Your presence score, computed entirely on your own device. Objective signals only, never an emotional read — and if there is not enough coverage to score it honestly, it returns nothing rather than a guess.

Gamification Level

Dashboard → momentum & level

Your level, earned only from real completed sessions across your whole history. No participation points.

Spaced repetition uses the ease-factor update from Algorithm SM-2, © SuperMemo World, 1991–2026, supermemo.com — our interval rule is our own and is not the canonical SuperMemo product. All other algorithms above are open, public methods with no license required. Sources on /licenses.

Also under the hood

EWMA current formOLS momentum±1σ projection bandstd-dev consistencycoefficient of variationrecency decaytype-token ratiopercentile vs. own historymedian baseline calibration

Anyone can wrap an LLM in a chat box. This algorithm layer is the hard part to copy — and the part that actually moves your score. It runs on data you already have, at zero extra cost.

Capabilities · 6 core features

What we build for you

AI question engine01

Tailored, never generic

Powered by a multi-model AI engine (Gemini on Premium) — every question is built from your role, level, and target company. Zero copy-paste prompts.

5 interview types02

A rubric for every format

Behavioral, Technical, System Design, Culture Fit, and Quant Finance — each scored on its own dimensions, in 8 languages.

Real-time feedback03

Scored the moment you answer

Instant scoring across structure, clarity, depth, and examples — no waiting for a report at the end.

Performance analytics04

Your trajectory, dimension by dimension

Every session tracked across every score dimension, so you always know exactly where the next point comes from.

Score cards05

Shareable, from a real score

A clean, LinkedIn-ready card generated from your actual result — never a fabricated number.

Company prep · Premium06

Tuned to one firm at a time

Google, Amazon, Meta — or any company you name. Question style and probing tuned to that firm's real loop.

Process · 4 steps

How it works

01
Step 1Create your profile

Set your target

Pick your role, experience level, language, and (on Premium) a target company. The AI calibrates question difficulty and scoring to match real hiring bars.

RoleDifficultyCompany target
02
Step 2Pick a session

Choose your mode

Five specialized interview formats — behavioral, technical, system design, culture fit, and quant finance. Each uses its own scoring rubric. Quick or coached mode.

BehavioralTechnicalQuant
03
Step 3Interview with AI

Practice with AI

A real interview powered by a multi-model AI engine (Gemini on Premium). Answer in text or speak out loud — Coached mode adds live delivery feedback on pace, fillers, and hedging.

Real timeVoice modeDelivery coaching
04
Step 4Review your results

Get your score

A detailed breakdown across communication, depth, structure, and examples — plus the exact model answer and what to fix. Download a polished score card to share anywhere.

Score breakdownModel answerShare card
Trust · 6 verified claims

Data you can trust.

No dark patterns, no overclaiming. Here is exactly how your account and your data are handled — every line below is true of how the product is actually built.

Sign-in

No passwords to leak

Authentication is handled by Clerk. We never store your password — there is nothing on our side to leak. Sessions are signed RS256 tokens, verified on every request.

Payments

Cards never touch us

Every payment is processed entirely by Stripe. Your card number never reaches our servers — we only ever see a subscription status.

Edge infra

Server-side by default

Everything runs on Cloudflare’s global edge over TLS. Every plan check happens server-side, and per-user rate limits guard the API from abuse.

Data ownership

Delete any session, permanently

Delete any interview from your history and it is gone — a real database delete, not a soft flag. We do not show your data to other users and never fabricate activity on your account. (Full account deletion is on the roadmap; today, deletion is per-session.)

AI training

Your answers don’t become someone else’s training data

Every AI engine we route your answers through — across every plan — contractually commits not to train on your inputs or outputs. Live-voice audio runs on Gemini’s paid tier, carrying the same no-training guarantee. Voice transcription is sent with training explicitly opted out on every request.

On-device presence

The camera feed never leaves your browser

Presence coaching runs entirely client-side — a face-landmark model loaded into your tab, measuring objective signals only (eye-contact %, posture stability) and never interpreting your emotions or behavior. Only those derived numbers are ever sent to our server; the video frames themselves are read locally and discarded. (The detection library does report anonymous performance telemetry to its vendor — never your video.)

What we won’t claim

  • —Full one-click account deletion doesn’t exist yet. You can permanently delete any individual interview today; deleting your whole account is on the roadmap.
  • —Gemini’s paid tier commits to not training on your live-voice audio or resume PDFs — it does not separately guarantee no human ever reviews flagged content for abuse detection.
  • —“Hedging” in your delivery score is purely a language metric — it counts hedging qualifiers (“maybe”, “sort of”) in your transcript, alongside a separate filler-word count. It scores the words you said, and makes no judgement about you as a person.

The honesty pledge

Every number on this site is a real capability or a value computed from your own sessions.

No fabricated user counts. No fake “hired at” logos. No invented match percentages. If we can’t measure it, we don’t show it.

Every AI provider, open-source license and algorithm source cited above is listed in full on /licenses.

Product tour
38 seconds · Sound on

See it run.

How it works, shot like a launch film — pick your target, answer like it's real, get scored in seconds, and watch your command center climb.

1 · Target2 · Answer3 · Score4 · TrackTrain under pressure

Real reviews

Reviews, not testimonials.

No hand-picked quotes. Vrenic’s users review it on its own page, where every published review is shown, newest first.

  • Verified use

    Only people who have completed at least 2 practice sessions can review, and each review shows how many they did.

  • Every rating

    Good or bad, every review is checked against the same published rules. A review is never removed for being negative.

  • No rewards, no insiders

    Nobody is paid for a review, the Vrenic team never posts one, and anyone can report one.

Ready to start

Land your dream role

3 free sessions. No credit card. Start in 60 seconds.

Begin sessionExplore features
A
M
J
K

5 types · 8 languages · any company

Free · No card · Start in seconds