Secondo Voice – Your AI Interviewer

Secondo Voice is an experimental app designed to delegate one-on-one interviews to an AI agent. The goal: free up interviewers' time by letting an AI conduct conversations anytime, in any language—and automatically generate relevant scores and summaries for each participant.

It’s especially useful for:

• Pre-screening in recruitment or casting
• Conducting international user interviews
• Automating any repetitive assessment workflow

---

🚀 Project Goal

This project was an experiment to build an Agent-as-a-Service during my free time, using modern AI tools to maximize development speed.

I documented my process and learnings in a blog post: 👉 Read the article

Built in January 2025 – some techniques may already be outdated, given the fast pace of AI development.

🔌 Why It's Offline

Secondo Voice was a proof-of-concept and personal productivity test.

Hosting the platform, especially the real-time audio services via Livekit—became too costly, so the project is no longer running publicly.

The front-end is still accessible here.

---

🧠 AI Agents In Action

The service relies on 6 specialized agents working together:

1. Voice Agent – Conducts the interview via speech (OpenAI).
2. Transcriber 1 – Converts audio to text using Whisper.
3. Transcriber 2 – A second pass to improve transcription accuracy using deepgram
4. Transcript Unifier – Merges the two transcripts into one clean version. (OpenAI)
5. Summarizer – Generates a summary of the conversation. (OpenAI)
6. Scorer – Evaluates answers based on predefined expectations. (OpenAI)

---

🤝 Co-Developers

• GitHub Copilot – A huge time-saver for the back-end API development.
• Bolt.new – Automatically generated the landing page and front-end UI based on constraints I provided (i18n, component architecture, etc.).

---

🛠 Tech Stack

• Database: Supabase – used for auth, CRONs, CDC, and more.
• ORM: Prisma – Hopefully one day we’ll have something this nice well maintained for Python.
• Back-end: FastAPI – Great for building REST APIs quickly.
• Front-end: React – The go-to library for dynamic user interfaces.
• Landing Page: Astro – Lightweight and modern.
• WebRTC: Livekit – Powerful real-time audio integration.
• Hosting: Render
• Others : Stripe, Sentry, Posthog, Umami