Demo mode (no server needed)
Sessions
Connected
Total sessions
All time
Active now
In progress
Completed
Ended
High risk
Signal count ≥ 5
All sessions
CandidateDomainLevelPhaseScoreQuestionsRiskTrajectoryStarted
Select a session to inspect
CandidateDomainScoreSignal countIntegritySmooth talkerTrajectory
Total API calls
— success · — failed
Avg latency
across all steps
API spend
LLM + STT + TTS
Success rate
Step breakdown
Latency by step — p50 / p95
Cost distribution by step
Call volume by step
Recent errors
Raw call log
TimeSessionStepModelLatencyTokensCostStatus
Completed sessions — select to review
CandidateDomainScoreQuestionsReviews doneStatus
Voice Settings
Speech-to-Text (STT)
Text-to-Speech
When disabled, browser speechSynthesis is used as fallback
Voice Provider
Voice
Clone Your Voice (Inworld API only)
Upload 30-120 seconds of clear speech audio. Uses Inworld instant voice cloning.
TTS Playground
Test text
Voice Library
Loading...
STT Playground
Record audio and test transcription across all STT providers
Provider
0:00
Prompt Playground
Edit the system prompt on the left. Test conversations in the chat on the right.
Model
Temp
Tokens
Candidate Variables
Name
Domain
Level
Tools
Years Experience
Question (for eval testing)
Answer (for eval testing)
System Prompt (sent with every message in chat)
Single Prompt (one-shot test)
Chat Test
Multi-turn conversation with system prompt
Set a system prompt on the left, then start chatting here.
Anti-Cheat Features
Enable or disable individual anti-cheat features. Changes apply to new sessions. Togglable features show a switch — always-on features are shown for reference.
Model Selection
Choose which LLM handles question generation and answer evaluation. Changes apply to new sessions immediately.
Question Generation
Needs speed over accuracy
Answer Evaluation
Needs accuracy over speed
Model Reference
Fast Tier
Haiku 4.5 — ~0.5-1s
Best for question generation
Balanced Tier
Sonnet 4.5/4.6 — ~1-2s
Best for evaluation
Premium Tier
Opus 4.6/4.7 — ~3-5s
Highest accuracy, slowest
Evaluation Prompt (editable — applies to next answer)
Evaluation Prompt Template
— runs once at interview end (after ≥8 answers). Variables: {name} {domain} {level} {years} {num_answers} {transcript}
Interviewer System Prompts
9 system prompts (3 domains × 3 levels). Select domain and level to view the prompt used by Ranjitha for that combination.
Domain
Level
prompts/trained_fresher_physical_design.md