Awaaz HR AI Hiring Infrastructure

The four-stage funnel

• Stage 1 — CV screen with scoring rubric + retrieval (RAG)
• Stage 2 — MCQ assessment with anti-cheat
• Stage 3 — Written assessment with plagiarism and copilot detection
• Stage 4 — Live voice interview conducted by an LLM agent

Six product surfaces

• Pipeline builder — compose any funnel as a directed graph; versioned, audit-logged, dry-runnable
• Voice agents — screener, interviewer, and scorer agents
• Automations — plain-English/visual triggers with idempotent retries and a dead-letter queue
• Configuration — rubrics as versioned, reviewable JSON
• Anti-cheat — tab-switch, paste, secondary-device, voice-deepfake, and copilot detection
• Cost ledger — every token, transcript second, and SMS attributed to a candidate

Governance & trust ledger

• Every write is signed (SHA-256 + hash chain); every action attributed to a user
• Reasoned impersonation with required reason and auto-expiring scope; read-only audit log
• GDPR-aligned, Postgres row-level isolation per tenant

Production telemetry & integrations

• 14,982 candidates/cycle, 11d 04h median funnel time, 8,407 LLM voice interviews to date, 99.96% p99 uptime
• Anti-cheat flags on just 0.41% of MCQ sessions
• 20+ maintained adapters: OpenAI, Anthropic, ElevenLabs, Twilio, LiveKit, Clerk, Supabase, Qdrant, Redis, Salesforce, HubSpot, Zapier, n8n, Make, Sentry, Prometheus
• Signed outbound webhooks (HMAC) with replay protection; typed TypeScript SDK and REST API inbound