Requirement Coverage

Every PDF requirement, mapped.

Source: Sales Agent Document — 230426.pdf. Each functional and technical requirement from the client document is mapped to a proposed component in the platform architecture.

Functional Requirements

PDF Requirement
Architecture Coverage
Status
Voice-only multilingual MVP
Real-Time Voice Layer with telephony gateway, STT, language detection, interruption handling, and TTS. Supports 9 Indian languages.
In Scope
Order taking
LLM structured extraction, order draft creation, validation engine, dealer confirmation loop, and Salesforce submission.
In Scope
Campaign manager
Business Control Plane with campaign console, campaign orchestrator, call queue management, retry rules, and suppression logic.
In Scope
Dealer profiling
Dealer Profile Store and Dealer Memory using category, market, depot, territory, prior call history, objections, and buying behavior.
In Scope
Scheme management
Scheme Engine, RAG Knowledge Layer, qualification rules, real-time balance lookup, eligibility criteria, and recommended pitch logic.
In Scope
Context, memory, and tonality
Conversation Policy Engine for context continuity, Dealer Memory spanning current and past interactions, and Tonality Controller with three modes.
In Scope
Salesforce + MCC/Oracle integrations
Enterprise Integration Layer for order punching, dealer detail lookup (Salesforce) and master data, scheme, catalog, backend sync (MCC/Oracle).
In Scope
Governance, reporting, and feedback
Audit Store for immutable event capture, Analytics Event Stream, management dashboards, and feedback workflow for AI improvement.
In Scope
Future scope
Extension hooks designed into the architecture: Salesforce case management, live agent transfer, colloquial voice tuning, and scheme follow-up engine.
Reserved

Technical Requirements

Infrastructure and system-level requirements from the source document.

Voice-only MVP channel
LiveKit-based voice runtime with telephony gateway. No text or chat channel in scope for MVP.
Covered
STT and TTS for Indian languages
STT with language auto-detection, noise handling, and IVR intrusion tolerance. Multilingual TTS for 9 languages.
Covered
Agentic AI layer
Full Agentic AI Brain: audience management, conversation flow, rule execution, structured extraction, tonality, memory, and analytics.
Covered
RAG for scheme and objection handling
Dedicated RAG Knowledge Layer with scheme rules, FAQs, and policy-approved objection answers. Updated via feedback loop.
Covered
LLM for structured output
LLM Orchestrator converts unstructured conversation into structured intent, objection, and order fields.
Covered
Self-hostable LLM option
Architecture supports private Llama 3 or equivalent model deployment alongside hosted model. Switchable per policy.
Supported
Salesforce integration
Integration Layer with Salesforce adapter for order submission, confirmation feedback, and dealer detail lookup.
Covered
MCC / Oracle integration
Integration Layer with MCC/Oracle adapter for master data sync, scheme data, product catalog, and balance feeds.
Covered

Governance Requirements

📄

Transcript Capture

Full verbatim transcript for every conversation, stored with call metadata and recording reference.

🛠️

Tool Call Audit

Every AI tool invocation — input, output, model context — recorded in the immutable audit store.

🎯

Scheme Recommendation Log

Scheme pitched per dealer, eligibility context used, and recommendation reasoning captured.

📋

Order Draft History

All draft states, correction events, confirmation, and final submission outcome recorded.

🔗

Integration Event Log

Every Salesforce and MCC/Oracle request and response stored with correlation IDs.

📊

Campaign Decision Audit

Dealer selection criteria, call queue decisions, retry logic, and suppression events recorded.

📈

Reporting Dashboards

Real-time views for call outcomes, campaign performance, agent performance, and order conversion.

🔄

Feedback Loop

Structured workflow to improve prompts, objection-handling rules, scheme knowledge, and campaign targeting.