Building NEONPLUS: My Journey Through the 10 Days of AI Voice Agents Challenge

Over the past ten days, the author participated in the 10 Days of AI Voice Agents (VoiceForBharat Edition) challenge. The objective was to build a production-ready, highly responsive voice agent for Indian users from scratch. Throughout the journey, all voice components were powered by Murf Falcon, utilizing expressive voices like Anisha, Samar, and Pooja. The result is NEONPLUS, a localized voice assistance system.

The main challenge was setting up the call analytics dashboard and testing local endpoints, which caused issues with the frontend interface on localhost:3000. The author resolved this by addressing port bindings and routing rules, clearing a local port conflict, and successfully hosting the SQLite-backed analytics dashboard.

To build a voice agent similar to NEONPLUS, follow these steps:

1. Speech-to-Text (STT): Use Deepgram configured for multi-language detection.

2. Language Model (LLM): Use Google Gemini for processing logic and conversation flow.

3. Text-to-Speech (TTS): Use Murf Falcon for low-latency, hyper-realistic voice output.

4. Real-Time Transport: Use LiveKit Agents framework.

The repository contains a practical guide, including code snippets, environment setup, and documentation. The author expresses gratitude to Murf AI, LiveKit, Deepgram, and Google Gemini for organizing the challenge and providing powerful tools for conversational AI.