Interview Simulator
Run a realistic, candidate-controlled mock interview. The simulator creates useful rehearsal conditions and evidence-linked coaching; it does not observe, infer, or judge a person's emotions, attention, integrity, or employability.
When to Use
Use for a timed coding, design, behavioral, presentation, deep-dive, or hiring-manager rehearsal; for voice or whiteboard practice; and for a debrief that connects feedback to the candidate's own recorded work.
Do not use for selecting, ranking, screening, or monitoring people; detecting gaze, emotion, confidence, fatigue, distraction, or deception; proctoring; or producing a prediction about interview or hiring outcomes.
Core Process
flowchart TD
A[Candidate selects goal, modality, and consent] --> B[Configure prompt and rubric]
B --> C[Run practice with candidate controls]
C --> D[Capture only consented, observable artifacts]
D --> E{Enough evidence for each rubric finding?}
E -->|Yes| F[Link finding to session evidence]
E -->|No| G[Mark feedback unavailable or uncertain]
F --> H[Candidate-led debrief]
G --> H
H --> I[Candidate accepts, edits, or declines next practice step]
1. Establish the practice contract
Before presenting a prompt, confirm:
- the round type, session goal, duration, modality, and starting difficulty;
- whether the candidate wants recording, a transcript, or retained whiteboard/code artifacts;
- any accessibility needs, breaks, pacing, and preferred feedback style; and
- that all feedback is for practice, not a hiring assessment.
Do not turn on a camera, capture background activity, infer attention, or inspect a second monitor. If a requested feature would require that information, explain that the simulator does not provide it and offer a candidate-controlled alternative such as a timed text prompt or self-recorded answer.
2. Run one bounded round
Use the round routing below to select a prompt and observable rubric. State the timebox and any materials allowed. The candidate may pause, request clarification, switch modality, reduce or raise difficulty, or end the session at any time.
For a difficulty change, ask first. A valid rationale is observable practice evidence such as "you completed the stated constraint and asked for a harder follow-up" or the candidate's explicit request. Silence, speech rate, camera behavior, and facial expression are not difficulty signals.
3. Capture evidence, not impressions
Record only artifacts the candidate elected to retain: prompt versions, response text, code, whiteboard snapshots, transcript segments, elapsed time, and candidate self-notes. For each rubric dimension, cite the relevant artifact or mark it unavailable.
Use language such as "the response named a cache invalidation strategy at 12:40" rather than "you seemed confident." Where artifact coverage is incomplete, say what cannot be assessed.
4. Debrief into an experiment
Produce at most three strengths, three improvement experiments, and one next-session proposal. Each item must include:
| Field | Requirement |
|---|
| Observation | A direct pointer to a consented session artifact. |
| Rubric | The round-specific criterion being applied. |
| Confidence | High only when the artifact directly supports the finding; otherwise limited or unavailable. |
| Experiment | A candidate-actionable change for the next rehearsal. |
The candidate can correct the record, remove retained artifacts where the host product supports removal, and accept, change, or decline the proposed next step.
Round Routing
| Selected round | Prompt and rubric source |
|---|
| Coding | senior-coding-interview |
| General SWE system design | Candidate-selected design prompt and agreed observable criteria; no general-design specialist is installed and ML material is not a default. |
| ML system design | ml-system-design-interview, only when explicitly selected. |
| Presentation | tech-presentation-interview |
| Behavioral or values | values-behavioral-interview |
| Hiring-manager conversation | hiring-manager-deep-dive |
| Technical deep dive | Candidate-selected project and factual project evidence; use anthropic-technical-deep-dive only for explicitly selected Anthropic research. |
A paired specialist supplies practice material within this session contract. Do not adopt employer predictions, inferred traits, or unverified process claims from specialist material. The candidate approves any target-specific scope and prompt.
Bundled References
Load references/INDEX.md for optional mobile practice or whiteboard implementation guidance. Both guides inherit the session contract, including consent before external processing and candidate approval before plan changes.
Difficulty and Modality Decisions
| Situation | Move |
|---|
| Candidate requests a harder or easier prompt | Change difficulty and record the request. |
| Candidate completes a stated constraint and asks for a follow-up | Offer one harder follow-up; do not force it. |
| Artifact is missing or transcript quality is poor | Keep the practice, label feedback limits, and offer a retry. |
| Voice connection degrades | Offer text mode, a local candidate recording, or a reschedule; do not score missing portions. |
| Whiteboard is inaccessible or unwieldy | Use verbal structure, typed outlines, or a shared text canvas chosen by the candidate. |
| Candidate asks for a realism check | Compare the configured format to publicly available process information, clearly separating known facts from assumptions. |
Quality Gates
Mark a session complete only when:
Anti-Patterns
Proctoring Disguised as Practice
Novice: "A realistic mock should monitor eyes, tabs, cameras, and second screens."
Expert: Practice is valuable when the candidate controls materials and boundaries. Gaze and device monitoring create privacy risk and do not establish interview quality or integrity.
Detection: Any feature or feedback mentioning gaze, focus, distraction, camera monitoring, second monitors, or integrity flags.
Affect as a Score
Novice: "Hesitation, vocal tone, or facial expression reveal confidence and should adjust difficulty."
Expert: Those signals are ambiguous and can reflect accessibility, culture, connection quality, or ordinary thinking. Adjust only at the candidate's request or from explicit task evidence.
Detection: Feedback that attributes confidence, fatigue, engagement, or emotion without a candidate-authored statement.
Debrief as a Hiring Forecast
Novice: "A mock score predicts whether the candidate will pass."
Expert: A session is a bounded practice sample. Rubric feedback should expose what was observed, what was missing, and a next experiment—not an employability estimate.
Detection: Percentile, pass/fail, ranking, or hiring-probability claims.
Provider Lock-In
Novice: "Voice, transcription, and whiteboard features should be designed around one vendor."
Expert: Keep capture, transcription, whiteboard, and feedback behind replaceable adapters. A provider outage or accessibility constraint should degrade to candidate-selected text or exportable artifacts.
Detection: A session contract that assumes a named provider, requires a camera, or cannot fall back to text.
Worked Example
Setup: A staff backend engineer chooses a 45-minute system-design rehearsal, typed outline plus voice, difficulty 3, and transcript retention. Their goal is to practice stating tradeoffs before implementation detail.
During the session: At 16:10, the candidate writes two scale constraints and asks for a harder follow-up. The simulator offers one constraint about regional failures. At 29:00, the voice connection drops; the candidate switches to text and keeps the session going.
Debrief:
- Strength: The candidate named latency and consistency constraints before proposing components, with transcript references at 04:22 and 16:10.
- Improvement experiment: In the next rehearsal, reserve two minutes to state an explicit failure-mode plan before drawing the data flow.
- Limitation: No speech-quality feedback is provided for the portion completed in text; the transcript does not support it.