Voice AI sales training, built and in production in two months

Project goal

Client: Modern Voice (modernvoice.ai), United States. Created by insurance agents and former agency owners.

The founders wanted agents to meet realistic objections before real calls: to rehearse with AI buyers who answer by voice, and to get a score after each session. There was no product yet. We started from zero.

Solution

We designed the architecture and built all the software:

• A Spring Boot backend with a live WebSocket bridge between OpenAI Realtime (gpt-4o-realtime, whisper-1) and ElevenLabs streaming speech (eleven_flash_v2_5).
• AI buyer personas with different temperaments and difficulty levels, and scoring with feedback after every session.
• Call recordings and transcripts, upload of the agency's own scripts, team dashboards and a leaderboard.
• An Angular web app with ECharts analytics.
• Stripe subscriptions with reconciliation of payment events.
• AWS infrastructure, CI/CD and the marketing sites on Next.js.

Team: two developers and a tech lead. Time from start to production: two months. Launch: 1 July 2025.

Results

• Launched on 1 July 2025.
• 100+ agencies and 400+ users.
• Three people took it from zero to production in two months.
• Public monthly plans for agents and a custom plan for agencies.

Why it matters for you: If you plan a voice agent, we have already connected real-time speech recognition, a language model and streaming speech synthesis in a paid product.