┌SAGAR // AI SYSTEMS
PRODUCTION RUNTIME • v2.6.0┐
AI Systems & Autonomous Agent Architecture
STATE
A
PROBLEM.
SPARK
THE
AGENT.
SAGAR
SHIPS.
[ TAP ANYWHERE TO ENTER ]
ENTER SYSTEM→
02 / 17 · AI Automation & Voice AI

VaniFlow: Multilingual Voice AI Employee

AI voice employees that handle live customer phone calls in 3 languages with zero lag.

VaniFlow: Multilingual Voice AI Employee
AI Automation & Voice AI
Architecture & Overview

VaniFlow provisions autonomous AI voice employees that handle real phone calls without walkie-talkie delays. Customers can speak naturally in English, Hindi, Telugu, or switch languages mid-sentence. Built with Gemini 3.8 Live with sub-200ms latency, it handles live interruptions seamlessly, qualifies customer leads, and books appointments directly onto Google Calendar.

The Problem

Traditional IVR systems and conversational voice bots suffer from high latency, rigid single-language prompts, unnatural turn-taking, and lack of real-time interruption handling.

The Approach & Implementation

Architected a low-latency pipeline with 16kHz browser AudioWorklet streaming, a Go WebSocket bridge to Gemini 3.8 Live, native barge-in packet cancellation, and automated post-call classification using Gemini 3.8 Flash.

Measurable Outcome

Achieved sub-200ms bidirectional voice latency with natural human-like turn taking, seamless Indic code-switching, and automated CRM record synchronization.

Technologies & Skills
Gemini 3.8 LiveGemini 3.8 FlashGoWebSocketsAudioWorkletSQLiteTailwind CSS
Key Verification Metrics
Sub-200ms bidirectional voice latency via Gemini 3.8 Live
Multilingual code-switching (English, Hindi, Telugu)
True native barge-in with instantaneous buffer drop
Automated post-call CRM lead qualification
Project Availability
View Repository
Codebase showcase · Live deploy upon request