VoiceHire v1.0.0

VoiceHire v1.0.0

VoiceHire v1.0.0 is the first production-ready release of the AI-powered, resume-aware voice interview platform.

The platform now provides a complete interview lifecycle, including resume analysis, real-time voice transcription, adaptive question generation, asynchronous response evaluation, session recovery, and final interview reporting.

Highlights

• Complete resume-aware voice interview workflow
• Real-time audio communication through LiveKit Cloud
• Live speech transcription with Deepgram Nova-3 through LiveKit Inference
• Adaptive question generation with Gemini 2.5 Flash
• Configurable multi-phase interview architecture
• Coverage-based phase completion and maximum turn limits
• Confluent Cloud Kafka-based asynchronous evaluation pipeline
• Incremental turn and phase evaluation
• Supabase-managed PostgreSQL persistence
• Session recovery and interview resumption
• Final evaluation polling and structured report generation
• Production deployment with Docker Compose, Nginx, and HTTPS

Interview Workflow

VoiceHire conducts interviews through configurable phases.

The default workflow includes:

• Intro phase — maximum 2 turns
• Experience phase — maximum 3 turns

Each phase controls:

• Question generation
• Structured information extraction
• Session state updates
• Coverage evaluation
• Completion conditions

Additional phases can be implemented through the shared InterviewPhase contract and registered with the interview engine.

Asynchronous Evaluation

VoiceHire evaluates interview responses incrementally:

1. Each submitted answer publishes a turn-assessment event.
2. The turn assessment consumer evaluates and persists response scores.
3. Completing a phase publishes a phase-evaluation event.
4. The phase evaluation consumer waits for the required turn assessments, aggregates the phase results, and persists them.
5. Phase results are generated while the interview continues.
6. Final evaluation reuses completed phase results to reduce end-of-interview latency.

Infrastructure

• React 19
• TypeScript
• Vite
• FastAPI
• Python 3.11+
• Gemini 2.5 Flash
• LiveKit Cloud
• Deepgram Nova-3
• Confluent Cloud Kafka
• Supabase-managed PostgreSQL
• Docker Compose
• Nginx
• Certbot and Let’s Encrypt

Production Deployment

The production Docker Compose stack includes:

• FastAPI backend
• React and Nginx frontend
• LiveKit transcription worker
• Turn assessment consumer
• Phase evaluation consumer
• Certbot

The application supports:

• AWS EC2 deployment
• HTTPS certificates issued by Let’s Encrypt
• Managed Confluent Cloud Kafka
• Managed Supabase PostgreSQL
• External LiveKit Cloud rooms and inference

See README.md for configuration, local development, database initialization, Kafka topics, worker startup, and production deployment instructions.

Upgrade Notes

This release introduces the session_phase_evaluations table and the following Kafka topics:

• voicehire.turn-assessment.requested
• voicehire.turn-assessment.completed
• voicehire.phase-evaluation.requested
• voicehire.phase-evaluation.completed
• voicehire.dead-letter

Existing deployments upgrading from VoiceHire v0.2.0 should:

1. Update the Supabase schema.
2. Create the required Confluent Cloud Kafka topics.
3. Configure the new Kafka and LiveKit environment variables.
4. Start the LiveKit transcription worker.
5. Start the turn assessment and phase evaluation consumers.
6. Rebuild and restart the Docker Compose stack.