Ten purpose-built modules.

Voice Calling Layer

Outbound calling, multilingual STT/TTS, interruption handling, and call recording. Supports English, Hindi, Bengali, Gujarati, Marathi, Tamil, Telugu, Kannada, and Malayalam.

• — LiveKit / SIP / PSTN telephony gateway
• — Noise and IVR-aware STT with language auto-detection
• — Natural multilingual TTS voice response
• — Call recording with reference storage

Agentic AI Brain

Conversation policy, campaign objective management, dealer context handling, scheme pitch, objection handling, and dynamic tonality selection.

• — LLM orchestrator with hosted model and optional private Llama 3
• — Conversation policy engine with guardrails and compliance rules
• — Standard, assertive, and persuasive tonality modes
• — Per-dealer memory spanning current and historical calls

LLM + Extraction Layer

Converts unstructured dealer speech into structured intent, objection, and order fields. Can start with a hosted model and later migrate to a private Llama-style deployment.

• — Real-time intent classification from natural speech
• — Objection type detection and routing
• — Structured order field extraction
• — Self-hostable LLM option for data residency requirements

Campaign Manager

Audience selection, campaign scheduling, dealer prioritization, call attempt management, suppression rules, and real-time call status tracking.

• — Dealer audience selection by market, category, scheme relevance
• — Campaign scheduling and priority queuing
• — Call attempt tracking with configurable retry rules
• — Suppression to prevent over-calling

Dealer Intelligence

Dealer category, market, depot, territory, previous calls, prior objections, and buying behavior — unified into a sales-specific intelligence layer.

• — Consolidated dealer profile across all channels
• — Previous call outcomes and objection history
• — Market, depot, and territory segmentation
• — Next-best-action context loaded at call start

Scheme Engine

Scheme eligibility rules, qualification criteria, current dealer balance, and recommended pitch generation based on dealer profile and campaign objective.

• — Scheme metadata ingestion from MCC/Oracle or managed upload
• — Real-time eligibility and balance lookup per dealer
• — Qualification progress tracking
• — Recommended scheme pitch with personalization logic

Order Engine

Product and quantity capture from natural speech, dealer confirmation loop, validation against catalog and scheme constraints, and Salesforce order submission.

• — Structured order draft from free-form conversation
• — Product and quantity confirmation read-back loop
• — Validation against catalog, scheme, and business rules
• — Salesforce submission with staging and retry on failure

Integration Layer

Salesforce integration for order submission and dealer detail lookup. MCC/Oracle integration for master data, scheme data, balances, and backend sync.

• — Salesforce order punching and status feedback
• — MCC/Oracle dealer, product, and scheme data sync
• — Internal adapters with staging tables for early builds
• — Fully auditable request and response logging

Governance Layer

Full audit trails, conversation transcripts, dashboards for call outcomes and campaign performance, and a feedback loop for improving AI behavior.

• — Immutable audit store for every call and tool action
• — Analytics event stream for campaign and order funnel
• — Management dashboards: conversion, outcomes, performance
• — Feedback workflow for prompts, rules, and objection handling

Salesforce Case Management

Identify and register dealer complaints or escalations directly into Salesforce cases during or after a voice call.

Live Agent Transfer

Hand off complex objections or high-value opportunities to a human agent mid-call, with full context transfer.

Colloquial Voice Tuning

Improve local and natural speech patterns specific to each language region — dialects, idioms, and informal expressions.

Follow-Up Engine

Automatically nudge dealers about scheme balance thresholds, pending actions, or expiring qualification windows.