💸 MoneyOps: Intelligent Financial Orchestration

Status: Phase 5 Complete (Smart CRM & Market Intelligence) 🚀 Architecture: Distributed AI Microservices with Founder Intelligence Layer

MoneyOps is a next-generation financial platform that orchestrates AI Agents to handle invoicing, payments, CRM, and business intelligence via Voice and Chat interfaces. It leverages a "CEO Brain" architecture to go beyond simple CRUD operations.

---

🏗️ System Architecture

flowchart TB
    classDef default font-size:16px,padding:8px;
    classDef subgraphTitle font-size:18px,font-weight:bold;

    subgraph ClientLayer [Client Layer]
        direction LR
        WebApp[Web App - React]
        LiveKitVoice[Voice Interface - LiveKit]
    end

    subgraph APIGatewayLayer [API Gateway - Spring Boot]
        direction TB
        APIGateway[Cloud Gateway] --> RateLimiter[Redis Rate Limiter] & Auth[JWT / Clerk Validation]
    end

    subgraph VoiceServiceLayer [Voice Service - Python/LiveKit]
        direction LR
        LiveKitAgent[LiveKit Framework] --> VAD[Silero VAD] & STT[AssemblyAI/Groq] & TTS[Cartesia/Groq]
    end

    subgraph AIGatewayLayer [AI Gateway - FastAPI]
        direction TB
        Intent[Intent Classifier] --> AgentRouter[Agent Router] 
        AgentRouter --> AgentsBox[Agents: Finance, Sales, Compliance, Market, Orchestrator]
        AgentsBox --> SessionMgr[Session Manager - Redis]
    end

    subgraph BackendCoreLayer [Backend Core - Spring Boot]
        direction TB
        CoreServices[OrgAdmin, Invoice, Client, Onboarding Services]
        Intelligence[Founder Intelligence Layer]
        CoreServices -->|Lead Scoring| Intelligence
    end

    subgraph DataLayer [Data Layer]
        direction LR
        MongoDB[(MongoDB Atlas)]
        RedisCache[(Redis - Cache/Sessions)]
        KafkaBus{Kafka Event Bus}
    end

    subgraph ExternalAPIs [External Integrations]
        direction LR
        Groq[Groq LLM]
        Tavily[Tavily Search]
        LiveKitCloud[LiveKit RTC Cloud]
    end

    %% Optimized Cross-Connections to avoid spreading
    WebApp -->|HTTPS| APIGateway
    LiveKitVoice -->|WebRTC| LiveKitCloud <--> LiveKitAgent
    
    APIGateway --> BackendCoreLayer & AIGatewayLayer
    LiveKitAgent -->|Intercept| AIGatewayLayer
    
    AgentsBox -->|Synthesis| Groq
    AgentsBox -->|Search| Tavily
    
    BackendCoreLayer --> MongoDB & RedisCache & KafkaBus
    AIGatewayLayer --> RedisCache

Loading

---

🧠 Data Flow & Intelligence Pipeline

1. Transcript Interception: Speech is captured via LiveKit, transcribed (AssemblyAI/Groq), and intercepted by the Voice Service.
2. Intent Classification: The AI Gateway classifies input into Operational, Strategic, or Conversational intents using Llama 3.3.
3. Founder Enrichment: For actions like CLIENT_CREATE, the system automatically:
   • Infers industry and business type.
   • Calculates a Lead Score (CEO-level value assessment).
   • Generates Smart Notes for CRM (pitch angles, ESG compliance).
4. Backend Execution: Core services update MongoDB and broadcast events via Kafka.
5. CEO Summary: The Orchestrator synthesizes data from Finance, Market, and Compliance agents to provide an actionable voice response.

---

🚀 System Components

Service	Tech Stack	Description
Backend Core	Java 17, Spring Boot, MongoDB	REST API, Business Logic, Organization Isolation
API Gateway	Spring Cloud Gateway, Redis	Traffic Routing, Auth, Rate Limiting
AI Gateway	Python, FastAPI, Groq	multi-Agent Orchestration, Tool Registry, LLM Synthesis
Voice Service	Python, LiveKit Agents	Real-time Voice I/O with VAD-split utterance buffering
Frontend	React, LiveKit SDK	Intelligent Web Dashboard & Voice UI

---

🛠️ Technology Stack & Third-party APIs

• LLM: Groq (Llama 3.3 70B Versatile)
• Voice: LiveKit (Cloud), AssemblyAI (STT), Cartesia (TTS), Silero (VAD)
• Search: Tavily (Market Intelligence), SearchAPI
• Auth: Clerk (Identity & Organization Management)
• Database: MongoDB (Persistence), Redis (Context & Rate Limiting)
• Infrastructure: Kafka (Event Bus), Docker Compose

---

📚 Documentation

• Architecture Deep Dive: Original design document. (Legacy reference)
• Troubleshooting: Decisions on UUID representation and VAD split handling.
• Testing Results: Latest pipeline verification.