01
Brief the coach
Drop in the job description and your resume. InterviewOS extracts the competencies that actually decide the role and plans a structured interview around them — and around your claimed experience.
For macOS · Apple silicon 100% on-device Zero cloud calls
Interview like it's real.
Improve like it's science.
InterviewOS runs a spoken, face-to-face mock interview entirely on your Mac. A recruiter avatar asks, listens, and watches — then hands you evidence-linked feedback on what you said and how you delivered it. Your voice, camera, and documents never leave the machine.
No account. No subscription trapdoors. One 2.2 GB download, then fully offline.
Tell me about a time you turned a struggling project around.
Gazesteady
Pace142 wpm
Posturealigned
Evidence-linked
"…we cut churn 18% in one quarter…"
↳ maps to JD req: customer retention ownership
Whisper · local STT
Vision · 12 fps on-device
0 packets sent
LLAMA 3.2 · METAL-ACCELERATED◆
WHISPER SPEECH RECOGNITION◆
REAL-TIME FACE & POSE LANDMARKERS◆
NATIVE RUST CORE◆
APPLE SILICON◆
LOCAL SQLITE MEMORY◆
LLAMA 3.2 · METAL-ACCELERATED◆
WHISPER SPEECH RECOGNITION◆
REAL-TIME FACE & POSE LANDMARKERS◆
NATIVE RUST CORE◆
APPLE SILICON◆
LOCAL SQLITE MEMORY◆
Session protocol
From job description to
deliberate practice in minutes.
Every session is planned before the first word is spoken — so the interview itself feels instant, uninterrupted, and unnervingly real.
02
Interview, out loud
A recruiter avatar asks. You answer — speaking, not typing. Local Whisper transcribes in real time while the perception pipeline reads pace, gaze, and posture at ~12 frames per second.
03
Read the evidence
Your debrief cites its sources. Every observation links to a transcript line, a job requirement, or an observed signal. No vibes, no black-box verdicts — just what to change and why.
Perception stack
Three channels.
One picture of your delivery.
Interviews are multimodal. So is InterviewOS. A real-time perception stack fuses voice, face, and pose — computed frame by frame on Apple silicon, and surfaced as reflective signals, never verdicts.
Voice
On-device Whisper transcription with pace, pauses, and filler-word analysis — measured in words per minute, not opinions.
Face
A face landmarker tracks gaze steadiness and engagement — the quiet signals interviewers register before you finish your first sentence.
Pose
A pose landmarker reads posture and presence in real time — so "sit up straight" becomes a measurement, not a memory.
Camera is opt-in. The full interview works camera-off — vision signals are an instrument you choose to switch on.
face_landmarks: 478
pose_joints: 33
inference: on-device
On-device by design
Your rehearsal is nobody's data.
The entire runtime — a 3-billion-parameter language model, Whisper speech recognition, and the vision pipeline — executes on Apple silicon, inside the app. There is no server because there is nothing to send.
Your Mac
llama-3.2-3Breasoning · Metal
whisperspeech → text
vision pipelineface + pose · 12 fps
sqliteyour history, local
▲ everything above this line stays here
0
cloud calls, ever
0%
functional offline
0 GB
of intelligence, stored locally
0
accounts required
Grounded in interview science
Coaching built on the literature, not folklore.
Decades of industrial-organizational research say structured, competency-based interviews predict performance — and unstructured chats barely beat chance. InterviewOS is built on the structured side of that literature.
- Structured & competency-based by default. Question plans target the competencies extracted from the actual job description — situational and behavioral formats, in a deliberate sequence.
- Framework-aware feedback. Answers are analyzed against STAR, CARL, and PARADE structures, so coaching speaks the language hiring panels actually use.
- Evidence or it didn't happen. Every finding carries its source: a transcript snippet, a role requirement, or an observed signal. You can audit every claim the coach makes.
- No composite "hireability" score — by design. A single number flatters and misleads. InterviewOS reports independent dimensions so practice stays diagnostic, not demoralizing.
DEBRIEF · Q3 of 8
evidence-linked
Specificitystrong
"…we cut churn 18% in one quarter by rebuilding onboarding…"
Structure (STAR)result arrived late
Situation ran 40s before the task was named. Lead with the stakes.
Role alignmenton target
↳ cites JD requirement: "owns retention metrics end-to-end"
no composite score · by design
The product film
See a session end to end.
Three minutes: from dropping a job description to reading an evidence-linked debrief.
PRODUCT FILM · IN PRODUCTION
Premiering with the first access wave.
Questions, answered
Asked before you ask.
Does any of my data leave my Mac?
No. The language model, speech recognition, vision pipeline, and your session history all run and live on your Mac. There are no cloud calls, no telemetry, and no account. After the one-time model download, InterviewOS is fully functional offline — you can verify it with your network monitor of choice.
What do I need to run it?
An Apple silicon Mac (M-series) running macOS 12 or later. On first launch the app downloads ~2.2 GB of model weights once, into your local Application Support folder. Everything after that is local and instant.
Where do the questions come from?
From your target role — not a question bank. InterviewOS reads the job description and your resume, extracts the competencies that decide the role, and plans a structured interview around them before the session begins. That is also why the interview itself never stalls to "think."
Will it give me a hireability score?
Deliberately not. Composite scores hide more than they reveal and turn practice into anxiety. InterviewOS reports independent dimensions — specificity, structure, role alignment, delivery — and every observation is linked to evidence you can audit.
Do I have to turn my camera on?
No. The full interview works camera-off. If you opt in, posture and gaze appear as reflective signals — instruments, not judgments — and the video frames are processed and discarded on-device, never recorded or uploaded.
How do I get access?
Access is rolling out in waves while we harden the experience on a wide range of Macs. Request access below and you'll receive a download link for the notarized app as capacity opens.
Limited access waves
The room is waiting.
Request your seat.
Tell us who you are and what you're interviewing for. Invites go out in small waves — practitioners with an interview on the calendar go first.