One platform for fair, multilingual first-round interviews
From reception registration to kiosk interviews, AI evaluation and human review — every step configurable, every decision traceable.
A calm, consistent interviewer that speaks your candidates’ language
Questions are spoken aloud and displayed in the candidate’s own script. Candidates answer naturally by voice, with a typed fallback for accessibility — and never see a score.
- English, हिन्दी, Hinglish and ગુજરાતી live today
- Server or browser text-to-speech with the best available voice
- Clear states: AI Speaking, Listening, Processing, Next Question
- Automatic retry when audio is unclear — no penalty
किसी नाराज़ ग्राहक को आप कैसे शांत करेंगे?
Difficulty that adapts — inside the bounds you set
After every answer the engine recommends harder, same or easier. Difficulty moves within your min/max tiers, with one bounded follow-up per round for borderline answers.
- Round blueprints per competency module
- Selection reason recorded for every question
- Job-specific questions preferred automatically
- Time and question limits enforced server-side
Every score comes with its evidence
Answers are evaluated against the question rubric, expected answer, keywords and your knowledge base — returning component scores, verbatim evidence, strengths, gaps and confidence.
- Relevance, accuracy, clarity, depth and language components
- Difficulty-weighted module scores, reproducible formula
- Recommendation bands from your thresholds
- Reviewer can adjust any score with a note
Design assessments like a product, not a spreadsheet
Pick from nine competency modules, set the weight, question count, answer time and starting difficulty for each round. The builder won’t let you save weights that don’t total 100%.
- HR & Communication to Situational Judgment
- Per-round answer time and follow-up rules
- Pass and review thresholds per assessment
- Versioned — every session records the version used
+ 3 more modules
Run a room of kiosks from one screen
Assign candidates to rooms and PCs, watch live sessions, pause, resume or override a failed system check — while devices report heartbeats and readiness automatically.
- Access codes bound to a specific kiosk PC
- Camera, mic, speaker and network checks before start
- Focus loss, full-screen exit and disconnects flagged for review
- Resume exactly where a candidate left off after a drop
Rahul M. · हिन्दी
IN USE
Anjali P. · ગુજરાતી
IN USE
Imran K. · English
PAUSED
Ready for candidate
ONLINE
Sneha R. · Hinglish
IN USE
No heartbeat
OFFLINE
Know exactly which AI said what — and why
Prompt versions, providers, models, latency, tokens and cost are stored with every evaluation and report. Configuration and security changes land in a tamper-evident audit log.
- Provider-neutral: OpenAI, Anthropic and more with fallback
- Versioned prompts with draft → active workflow
- Monthly AI budget and per-session cost
- Secrets redacted from every log
Activated evaluator prompt v4
Neha (Assessment Mgr) CONFIGURATION
Rotated OpenAI key ••••7Q2c
Arjun (Super Admin) SECURITY
Evaluated Q12 via evaluator v4 · 0.9s
AI Engine AI
Adjusted score 62 → 70 with note
Kavita (Reviewer) REVIEW
Ground interviews in your products and policies
Upload product sheets, SOPs and FAQs. Evaluation retrieves the relevant passages, and question generation drafts new questions your team approves before use.
- Chunked, searchable documents per module or job
- Use for evaluation, question generation or both
- Generated questions start as drafts
- Keywords and red-flag terms per competency
Generated draft · awaiting approval
How is savings interest calculated, and when is it credited to the customer?
Calibrate the evaluator against your best reviewers
Score sample answers as humans, run calibration to compare with the AI, and promote the best examples to few-shot guidance — measuring agreement before you change anything in production.
- Human vs AI agreement metrics per prompt version
- Import reviewed session answers as samples
- Few-shot examples per competency
- Dataset export for offline analysis
Samples scored
120
Within ±10
91%
Few-shot examples
6
Illustrative example · x = human score, y = AI score
Everything a fair first-round interview needs
From the reception desk to the reviewer’s final note — one platform, configured by your team, auditable end to end.
Multilingual voice interviews
Candidates speak naturally in English, Hindi, Hinglish or Gujarati — questions are spoken and shown in their own script.
Adaptive T1–T4 questioning
Difficulty moves up or down within your bounds after every answer, with a recorded reason for each question selected.
Rubric-based evaluation
Every answer is scored against the same rubric — relevance, accuracy, clarity, depth and language — with evidence quotes.
Configurable modules & weights
Compose assessments from nine competency modules. Weights must total 100% — the score formula is always reproducible.
Secure rooms & proctoring
Dedicated kiosk PCs, room/device assignment, live operator controls and review-only proctoring signals.
Auditable AI
Prompt versions, models, providers, latency and cost are stored with every evaluation. Nothing is a black box.
Knowledge base grounding
Upload product and policy documents — evaluation and question generation are grounded in your own knowledge.
Train AI calibration
Score sample answers as humans, compare with the AI and promote the best examples to few-shot calibration.
See VaaniAssess in your assessment room
Book a 30-minute walkthrough. We'll run a live voice interview in the language of your choice.