AI Interview Room

Live mode with inline assistant (gpt-4o · Deepgram nova-2 · Vapi voice). Add NEXT_PUBLIC_VAPI_ASSISTANT_ID to .env.local to use a pre-created Vapi dashboard assistant instead.

Farrukh · AI Interview OS

Command Center

Live AI Interview · Senior ML Engineer

Irfan Malik·Senior ML Engineer·Round 2 of 3·INT-2026-04293

Idle

Excellent

00:00:00

Stage

AI InterviewerLive

AI interviewer

Farrukh · AI Interviewer

Voice model · Aurora-Neural v3

Idle

1080p · 60fps

Human Co-InterviewerNot joined

Priya Kapoor

Awaiting join

Mic mutedIdle

AM

Irfan Malik

Senior ML Engineer · Bengaluru, India

Idle

720p · 30fps

AI Activity

Listening

Listening

Transcribing

Analyzing

Generating next question

Speaking

Now:Capturing candidate audio in real time and waiting for natural speech boundary.

Live Transcript

Streaming

0 lines

No transcript matches your filters.

Interview Insights

Updating

84Candidate score

Communication

91

Confidence

79

Technical depth

86

Engagement

82

Recommendation

Strong Hire

Auto-updated by Farrukh rubric engine

Detected traits

Structured thinkingSystems mindsetCalm under pressureStrong ownershipPragmatic

Recruiter Notes

0 saved

No notes yet.

Recruiter Controls

Live actions · keyboard friendly

Session Summary Preview

Draft report

Overall Comm. Score

91

Composite weighted index

Top strengths

· Structured technical reasoning
· Strong systems mindset
· Clear, paced delivery

Top risks

· Light on calibration depth
· Brief pauses on cost questions
· Limited cross-team signals

Completion

0%

Elapsed · 00:00:00

Next action

Schedule live deep-dive with hiring panel

Suggested by Farrukh rubric engine

AI Settings

Post-interview analyser · OpenAI

Checking server-side key…

Enable AI analysis

When off, the result screen uses the local mock analyser.

Using local mock analyser

Add OPENAI_API_KEY to .env.local to enable.

Model

Pick from curated list or supply a custom OpenAI model id.

Sampling

Tune randomness and length for the analysis output.

Temperature0.40

0 = deterministic, 2 = wild. 0.3 — 0.6 works well for analysis.

Top-p1.00

Nucleus sampling. Leave at 1 unless you know you need it.

Max output tokens

Cap for the JSON response. Larger = longer reports, more cost.

Response format

OpenAI structured output mode.

System prompt

Editable. Sent verbatim as the system message.

Hidden by default. Reveal to fine-tune the analyser persona, rubric, or output schema.

Test connection

Pings the model with a tiny prompt and reports latency.

Settings are saved to your browser (localStorage). The OpenAI key stays on the server.

Prompt Studio

ML Platform Engineer · 8 prompts · 1 job

Job

Warm intro & rapport

hrfriendlyIntroduction

Greet the candidate warmly, confirm their background, and invite a brief self introduction.

Expected quality

60

End-to-end ML system design

technicalprobingTechnical RoundFollow-up

Ask the candidate to walk through a recent production ML system they owned, end to end.

Expected quality

85

Latency & scaling tradeoffs

technicalprobingTechnical RoundFollow-up

Probe how the candidate balances inference latency, freshness and cost.

Expected quality

82

Cross-functional disagreement

behavioralformalBehavioral RoundFollow-up

Ask about a time the candidate disagreed with a stakeholder and how they resolved it.

Expected quality

78

Mentorship & ownership

behavioralfriendlyBehavioral Round

Explore moments where the candidate mentored peers or took ownership of ambiguous work.

Expected quality

75

Culture fit signals

culturefriendlyBehavioral Round

Ask the candidate what kind of team environment lets them do their best work.

Expected quality

70

Hypothetical incident

situationalstrictTechnical RoundFollow-up

Pose a production incident scenario and observe the candidate's structured response.

Expected quality

88

Final pitch & questions

hrformalFinal Assessment

Invite the candidate to ask their own questions and summarize why they're a fit.

Expected quality

72