RiverbornBook Call
Dhoni AIReal-time Voice Intelligence

AI voice that listens,
responds, and remembers.

A real-time conversational voice engine that transcribes, understands, responds, speaks, and acts—all inside a single live call. Deployable on mobile, web, or telephony infrastructure.
Powering Vocalo.ai in production, serving 20,000+ users.

TELEPHONY STREAM STREAMER // INBOUND SIP-TRUNK

Connection: Twilio Media Stream WebSocket (ws://api.riverborn.com/v1/voice/stream)
01 / Telephony Ingest
Inbound WebSocket Stream
Live Ingestion Transcript:
"I want to book a service appointment for my sedan next Tuesday morning."
02 / AI Engine Orchestration
STT ➔ LLM ➔ TTS Loop
[STT] Speech-to-Text150ms
[LLM] Agent Router1000ms
[TTS] Text-to-Speech250ms
Orchestration Logs:
STT Engine Parsing Input Speech...
03 / Playback Buffer
Outbound Audio Stream
Synthesized Voice Output:
Generating speech stream...
SPEECH INGEST:150ms // STT
LLM AGENT LOOP:1000ms (Generally 1s)
VOICE SYNTHESIS:250ms // TTS
TOTAL LOOP LATENCY:1400ms (~1.4s Response)

Voice is still the hardest channel to automate well.

Most "voice AI" is either a brittle IVR tree that frustrates callers, or a voicebot that can talk but can't actually do anything—no bookings, no lookups, no follow-through. And almost none of it learns anything from the call after it ends.

Businesses that depend on phone volume, call centers, language platforms, and booking-heavy services are stuck choosing between a human team that doesn't scale and a bot that doesn't really help.

The Brittle IVR Tree

Traditional voice bots rely on rigid key-matching directories. They cannot handle interruptions, spelling corrections, or shifts in dialog context, leaving users stuck in circular phone menus.

Actionless Conversations

Existing speech agents cannot interact with database applications. They might chat but fail to execute live function calls, meaning bookings, account checkouts, or ticket edits still require human support agent handoffs.

Disconnected Post-Call Data

Most voice bots drop call content once the session hangs up. There is no automated analysis of the user's vocal performance, error patterns, or grammatical struggles to generate follow-up materials.

Live interactive call.
Post-call intelligence.

Dhoni runs a real-time conversational loop—hearing, responding, and executing live API calls—combined with background post-call pipelines that immediately analyze audio to deliver personalized follow-up.

Live Conversational Agent

Not a Branching Script

Dhoni operates as an intelligent voice agent capable of handling interruptions, tool lookups, and context management directly inside the call. By bypassing brittle IVR structures, the conversation flows naturally at human pace.

Post-Call Pipeline (Async)

Personalized Follow-up Content

Immediately after call completion, Dhoni's post-call webhook parses the audio and transcript to generate speech evaluations, compliance reports, and structured study exercises directly generated to address the exact gaps detected.

Powering Vocalo.ai

Vocalo is an AI language-learning platform where users have live spoken conversations with an AI agent to practice a language in real time.

During the call, Dhoni handles the full conversation naturally. After the call, it analyzes the transcript and audio to generate detailed, IELTS-style feedback, scoring and explaining exactly where the user struggled. Based on the specific mistakes detected, it then dynamically generates a personalized quiz so the user can drill the exact gaps in their ability.

20k+Sessions Served
< 1.4sLatency Loop
C1/C2AI Evaluation
Vocalo Session #48290-EN
OVERALL BAND
7.5IELTS
Level: Advanced (C1)
Grammar Range7.5

Complex structures used correctly, minor verb errors.

Vocabulary7.0

Good topic-specific terms, could use more idiomatic phrases.

Fluency & Coherence8.0

Fluid pacing, minimal hesitation, clear connectors.

Pronunciation7.5

Clear speech with minor vowel reduction on unstressed syllables.

PHONETIC & GRAMMATICAL ASSISTANCEDhoni AI Coach
Spoken

"Yesterday I have went to the store and buyed new shoes."

Correct

"Yesterday I went to the store and bought new shoes."

Reason: Use simple past tense ("went" / "bought") for finished actions in past time. Avoid combining present perfect "have" with irregular past tense "went".

Where Else This Fits

The same engine, the same real-time loop, applies anywhere a phone conversation needs to happen at volume and end in an outcome.

01

IVR & Call Centers

Replace static phone trees with a natural conversation agent that actually resolves the call, verifies credentials, and syncs account information.

02

Booking & Scheduling

The voice agent checks calendar availability, books appointments, and triggers confirmation alerts directly inside the call stream.

03

Healthcare Telemetry

Secure appointment booking, patient intake validation, and compliance reminders handled automatically with medical-grade reliability.

04

Banking & Insurance

Automate account queries, claims filing registration, policy FAQ routing, and security validations via live function-calling loops.

05

Automotive Services

Automate service booking, vehicle inventory lookups, diagnostic booking routing, and lead qualification via voice loop channels.

06

Language Practice

Live spoken conversations with assessment dashboards and personalized follow-up drills scaled without manual intervention.

Voice Engine Architecture

Dhoni runs two separate pipelines: a live low-latency media stream (in-call) and an asynchronous post-call intelligence webhook thread (2m latency) that syncs records and compiles evaluations.

01Conversational

Low-latency stream

Structured to handle turn-taking under 1.4s total roundtrip, utilizing streaming Speech-to-Text and fast Text-to-Speech models over WebSockets.

02Transactional

Live function calls

The agent executes external API calls mid-sentence, pulling records or booking calendars without breaking the conversational audio loop.

03Multi-channel

Channel-agnostic core

Telephony-ready engine architecture built to run natively over Twilio media stream gateways, web interfaces, or custom mobile voice protocols.

04Asynchronous

Post-call intelligence

A background analysis queue analyzes complete audio recordings for grammar, vocabulary, pronunciation, and automatically pushes logs to the CRM.

Engine Architecture Blueprint

Trace the live bidirectional real-time audio transport loop (in-call) and the async post-call analytics thread triggered via event webhooks.

PIPELINE A: REAL-TIME STREAM LOOP (IN-CALL CONVERSATIONAL AGENT)PIPELINE B: ASYNCHRONOUS POST-CALL ANALYTICS THREADTRANSPORTWebRTC / SIP / WSSTTDEEPGRAMLLMCLAUDE / GPTAPI CALLCALENDAR DBTTSCARTESIAWEBHOOKPOST-CALL HOOKPOST-CALL SESSION HOOKINGEST ENGINEAudio + Text IngestSPEECH COACHEvaluate AudioCRM / SMS OUTPersonalized Quiz
Active System Phase:System Idle
Diagnostics Logs:Waiting for incoming real-time audio transport to initialize session pipeline...
Telemetry:STANDBY
CRM SYNC COMPLETETotal Latency: 2m
Post-Call Thread FinishedIELTS assessment score & grammar gap quiz generated and pushed to Vocalo student portal.

Engagement Models

Deploy Dhoni AI voice intelligence pipelines into your stack according to your compliance and scale constraints.

Model 01

White-label SaaS

Run Dhoni AI under your own corporate brand, configured on your customer-facing product domain. Enforce your user account plans, storage rules, and frontend visual styles with ease.

Fully Hosted & Managed
Model 02

API Access

Integrate our voice engine pipelines directly into your software workflow, mobile applications, or legacy CMS. Submit raw voice recordings and fetch evaluations programmatically.

Low-latency REST / Webhooks
Model 03

Custom Vertical Build

A custom-trained conversation and function-calling layer engineered specifically for your business sector or industry context, built on top of our telephony core.

30-Day Custom Delivery

Proven voice loops

Live conversation. Real outcome on every call.

Deploy Dhoni AI speech pipelines into your business stack. Connect with our engineering team to audit your telephony infrastructure or test API integrations.