Enterprise Voice AI.
Built Beyond Call Centers.
Inno-vakya combines conversational AI, intelligent routing, omnichannel communication and real-time observability into one unified platform.
Customer
Inbound intent
Phone / WhatsApp
Voice · SIP · Chat
AI Agent
STT → LLM → TTS
Routing Engine
Skill-based · LAA
Human Agent
WebRTC handoff
Supervisor
Live oversight
Traditional contact centers break down at scale.
Disconnected tools, static routing and zero observability turn every customer conversation into a gamble. Inno-vakya was built to fix all six failure modes at once.
- 01
Long wait times
Callers sit in queues with no agents free, abandoning before they are ever heard.
- 02
Repeating information
Customers re-explain context at every transfer because nothing follows them.
- 03
Poor routing
Static IVRs send people to the wrong skill, then bounce them around.
- 04
Fragmented channels
Voice, WhatsApp and dialer each live in their own silo with separate logic.
- 05
Separate systems
Bot, telephony, CRM and reporting are stitched together and drift apart.
- 06
Limited visibility
Supervisors fly blind — no live quality, latency or routing telemetry.
Three pillars. One shared AI brain.
Voice, chat and routing intelligence are not bolted together — they run on a single persona, one database and one assignment pipeline.
Voice
Real-time conversational AI over PSTN & WebRTC
- Multilingual EN / HI / OR bot
- Interruptible, natural speech
- Inbound, outbound & campaign dialer
- Seamless human handoff
Chat
WhatsApp conversations on the same AI brain
- LLM persona adapted for text
- Full transcript on handoff
- Proactive overdue reminders
- Live agent chat inbox
Routing Intelligence
One race-free engine shared by every channel
- Skill queues on Redis Streams
- Longest Available Agent claim
- Priority requeue & timeout watcher
- Fault-tolerant recovery
Conversations that feel human, in real time.
A sub-second STT → LLM → TTS loop, multilingual and interruptible — with a clean path to a human whenever the customer needs one.
Multilingual AI
English, Hindi and Odia in a single persona.
Real-time conversations
Sub-second STT → LLM → TTS turn loop.
Interruptible speech
Barge-in support — callers can talk over the bot.
Natural TTS
Cartesia voices with human cadence and prosody.
Speech recognition
Sarvam STT tuned for Indian accents & code-mixing.
Context awareness
The LLM tracks intent and decides when to escalate.
Human handoff
route_to_agent hands a live call to the right skill.
Call recording
Vobiz recordings synced to S3 with presigned playback.
Voicemail
Auto-offered when waits exceed 5 minutes; stored to WAV.
Every channel, one routing brain.
Inbound voice, outbound dialer, automated campaigns and WhatsApp all flow through the same assignment pipeline — no silos, no duplicate logic.
Inbound Voice
PSTN calls land on the AI bot, queue by skill, then connect to an agent.
- Vobiz SIP → LiveKit
- Skill queueing
- Wait-time announcements
Outbound Dialer
Agents dial India numbers from a softphone with Indian-cadence ringback.
- Manual dialpad
- Live answer events
- SIP via Vobiz
Campaign Dialer
Upload an XLSX and the engine pairs idle agents with pending contacts.
- IST 9am–8pm window
- Stale-call recovery
- Auto context pop
WhatsApp AI
Twilio chat handled by the same brain, escalating with full history.
- LLM text persona
- handoff_to_agent
- Overdue reminders
A race-free engine that always finds the right agent.
One asyncio worker per skill group consumes Redis Streams; an atomic Lua claim implements Longest Available Agent in a single round-trip — with timeouts, requeues and crash recovery handled for you.
- 01
Skill queues
One asyncio worker per skill group — sales, servicing, outbound, chat.
- 02
Redis Streams
Escalations are enqueued as stream entries consumed by skill workers.
- 03
Longest Available Agent
An atomic Lua claim picks the longest-idle free agent in one round-trip.
- 04
Priority requeue
Rejected or timed-out calls jump to the head of the queue.
- 05
Reservation timeout
A sweeper reclaims reservations the agent never accepted within 30s.
- 06
Fault tolerance
Hangups, browser crashes and engine restarts all recover cleanly.
- 07
Agent assignment
A screen-pop lands in the agent's browser over a live WebSocket.
A softphone built for speed and focus.
Screen-pop assignments, a live call timer, wrap-up state and a chat inbox — everything an agent needs over a single reconnecting WebSocket.
Incoming Call
Priya Sharma
+91 98••• ••432
- Skill
- Servicing
- Intent
- Bike pickup reschedule
- Accept within
- 00:21
On Call
Connected
03:42
Wrap-up: 00:15Channels
WhatsApp · Rahul
“Can I extend my rental?”
Campaign · Renewals
Next: Anita K · context ready
Department-scoped control, end to end.
Manage agents, campaigns, recordings and voicemails — with live operational metrics and click-through drill-down into every conversation.
Live calls
12
Avg wait
0:38
Answer rate
94%
Call volume · last 12h
+18%- Agent management
- Department scoping
- Call recordings
- Voicemails
- Campaign dialer
- Chat history
Recording playback
Every call, measured. Nothing flies blind.
Prometheus + Grafana dashboards and a composite Voice Quality Score expose MOS, latency, jitter, packet loss, LLM latency and cost in real time.
Composite Voice Quality
Blends STT, LLM & TTS time-to-first-byte and turn lag (0–100).
One pipeline, from PSTN to dashboard.
Hover any service to see the role it plays. Three planes — voice, WhatsApp and control — share one Redis and one SQLite.
stage 01 / 10
PSTN
Public telephone network — where real customer calls originate.
Built on battle-tested foundations.
Best-in-class open and managed components, composed into one coherent, self-hostable platform.
- LiveKit
Media SFU / SIP
- Pipecat
Voice pipeline
- OpenAI
LLM
- Sarvam
Speech-to-text
- Cartesia
Text-to-speech
- FastAPI
API + WebSocket
- Redis
Routing state
- SQLite
Call logging
- Next.js
Frontend
- Docker
Packaging
- Prometheus
Metrics
- Grafana
Dashboards
- Twilio
WhatsApp
- Vobiz
SIP trunk
- AWS
Infrastructure
Everything you need to run a modern contact center.
- Multilingual AI
- Real-time Voice
- WhatsApp AI
- Skill Routing
- Campaign Dialer
- Voicemail
- Call Recording
- Supervisor Dashboard
- Observability
- WebRTC
- SIP
- LLM
- Open APIs
- Horizontal Scaling
- Natural TTS
Your data, your infrastructure, your control.
Inno-vakya is self-hosted by design — a single-tenant stack you run inside your own environment, hardened for enterprise.
Self Hosted
Runs entirely on your own infrastructure — no third-party cloud lock-in.
Private Deployment
Single-tenant stack inside your VPC or data center.
TLS Everywhere
Caddy-terminated TLS across every public surface.
Role-Based Access
Voice / Chat / Supervisor roles map to scoped permissions.
Audit Logs
Full call and chat audit trails for compliance review.
Secure APIs
Authenticated FastAPI surface with WebSocket auth.
Run it anywhere you operate.
From a single VM pilot to a horizontally-scaled cluster — the same stack, deployed to fit your environment.
Single VM
The whole stack on one EC2 instance for pilots.
Docker Compose
Voice/media stack containerized and reproducible.
Kubernetes
Scale routing workers horizontally on a cluster.
AWS / Azure
Deploy to your preferred public cloud region.
On-prem
Fully air-gapped installs for regulated workloads.
Hybrid Cloud
Split media and control planes across environments.
Build the Future of Customer Conversations.
See Inno-vakya handle a live voice and WhatsApp conversation on one AI brain — with routing, agent handoff and observability end to end.