All projects

AI · Telephony

Ongoing

1 March 2026

VoiceBot

Multi-tenant voice AI platform. Businesses deploy AI agents for inbound/outbound phone calls and web voice demos, with real-time STT/TTS in English and Indian languages.

Node.jsVue.jsPlivoDeepgramSarvam AIMySQLRedisKubernetes

VoiceBot is a multi-tenant platform for deploying AI voice agents. Businesses sign up, create an organisation, and configure agents. Each agent handles inbound calls, runs outbound campaigns, and can be embedded as a web widget.

What it does

Inbound calls: Agent picks up, transcribes speech in real time, generates a response via LLM, and speaks it back. Full transcript and recording saved automatically.

Outbound campaigns: Upload a CSV of contacts, select an agent, set a schedule. The platform dials each number automatically, runs the conversation, and retries failed calls.

Web voice demo: Embeddable widget powered by AudioWorklet + WebSocket. Visitors can try the agent directly in the browser.

Tech Stack

LayerTechnology
BackendNode.js + Express
FrontendVue.js 3 (dashboard + landing page)
TelephonyPlivo (primary) + Twilio
STTDeepgram (multi-language, low latency)
LLMGemini · OpenAI · Groq (provider-swappable)
TTSSarvam AI (Indian language synthesis)
DatabaseMySQL + Prisma ORM
Object storageMinIO
QueueBullMQ + Redis
OrchestrationKubernetes (k3s + Traefik)
CI/CDGitHub Actions + GHCR

Language support

LanguageStatus
English
Hindi
Marathi
Gujarati
MorePlanned

Language is configured per agent. No code changes required to add a new language once the AI provider supports it.

Provider abstraction

The AI provider is fully swappable via a single env var (VOICE_MODEL). Switching Gemini ↔ OpenAI requires zero code changes. Same pattern for telephony: the Twilio bridge is already built alongside Plivo.

Status

Core platform is working: inbound calls, outbound campaigns, STT/TTS, and the web widget. Productionisation, billing integration, and multi-language intent handling are ongoing.