Enterprise Agentic AI System (RAG) on Azure OpenAI | Chauffeur Dispatch Platform

Production-grade multi-agent AI system for luxury chauffeur booking, powered by Semantic Kernel, Azure AI Search (RAG), and real-time SQL-based availability.

Bravado Solutions Enterprise software development company building scalable AI systems, SaaS platforms, and cloud-native applications.

---

🚀 Overview

Most AI systems fail because they stop at conversation. This project demonstrates a real-world agentic AI system that goes beyond chat to:

• Understand user intent
• Retrieve enterprise knowledge (RAG)
• Check real-time availability
• Execute bookings with transactional safety
• Maintain conversational memory
• Operate as a scalable API service

---

🧠 What This System Does

An AI Concierge capable of:

1. Answering fleet, pricing, and service queries (RAG).
2. Checking real-time vehicle availability.
3. Booking chauffeur rides with atomic transactions.
4. Maintaining session-based conversations.
5. Persisting memory for context-aware responses.

---

🏗️ Architecture

This system follows an enterprise agentic architecture:

graph TD
    %% User Layer
    User((User)) -->|Booking / Info Request| API[FastAPI Orchestrator]

    subgraph "The Brain: Agentic Core"
        API --> SK[Semantic Kernel]
        SK <-->|Reasoning Loop| GPT[Azure OpenAI GPT-4o]
        SK <-->|Context Retrieval| Mem[Persistent Memory Store]
    end

    subgraph "The Hands: Plugin Layer"
        SK --> Plugins{Function Dispatcher}
        Plugins --> KP[Knowledge Plugin]
        Plugins --> BP[Booking Plugin]
        Plugins --> AP[Availability Plugin]
    end

    subgraph "Data & Knowledge"
        KP -->|Semantic Search| AIS[Azure AI Search]
        BP -->|SQL Transactions| DB[(SQLite Fleet DB)]
        AIS ---|1M+ Docs| Docs[Fleet, Pricing, Events]
    end

    %% Response Flow
    Plugins -->|Executed Action| SK
    SK -->|Final Answer| API
    API -->|Confirmation| User

    %% Styling
    style SK fill:#0078d4,stroke:#005a9e,color:#fff
    style GPT fill:#107c10,stroke:#094a09,color:#fff
    style Mem fill:#5c2d91,stroke:#3a1c5c,color:#fff
    style DB fill:#f29111,stroke:#b36b08,color:#fff

Loading

🔄 Core Flow

1. User Request → Initiated via FastAPI or Local CLI.
2. Orchestrator (Semantic Kernel) → Plans response, analyzes intent, and invokes tools.
3. Plugins (Tool Layer) → Knowledge (Azure AI Search), Availability (Fleet DB), and Booking logic.
4. Memory Layer → Stores and retrieves past interactions for long-term context.
5. Execution → Final response generated and action performed.

---

⚙️ Key Components

🔹 Orchestrator

• Central Reasoning Engine: Powered by Semantic Kernel to coordinate the model and tools.
• Task Planning: Handles intent recognition and dynamic tool selection.
• Execution Loop: Manages the flow between the LLM and plugin results.

🔹 Plugins (Tools)

• Knowledge Plugin: High-speed RAG via Azure AI Search for fleet and policy queries.
• Availability Plugin: Real-time queries to the Fleet Database for vehicle stock.
• Booking Plugin: Transactional logic to secure rides and update SQL state.

🔹 Memory

• Persistent Storage: Interaction history stored via Azure AI Search Vector Store.
• Context Retention: Maintains user preferences across multiple sessions.
• Extensible: Supports Redis, Pinecone, or other vector providers.

🔹 API Layer

• Framework: Built with FastAPI for high-concurrency async performance.
• Infrastructure: Includes session management, Redis rate limiting, and structured logging.

---

📂 Repository Structure

chauffeur-agentic-rag/
│
├── main.py                 # CLI entry point for local testing
├── app.py                  # FastAPI application entry point
├── .env.example            # Template for environment variables
├── requirements.txt        # Python dependencies
├── Dockerfile              # API container configuration
├── docker-compose.yml      # Multi-container orchestration (API + Redis)
│
├── scripts/
│   └── init_fleet_db.py    # Database schema and seed data setup
│
├── kernel/
│   └── builder.py          # Semantic Kernel initialization & configuration
│
├── plugins/
│   ├── knowledge_plugin.py # RAG & Azure AI Search logic
│   ├── booking_plugin.py   # Transactional ride booking operations
│   └── availability_plugin.py # Real-time fleet SQL queries
│
├── services/
│   └── orchestrator.py     # Core agentic reasoning & planning logic
│
├── memory/
│   └── vector_store.py     # Persistent context & vector search implementation
│
└── utils/
    └── pii_utils.py        # Data privacy and PII masking utilities
---

---

🏗️ Engineering Principles

• Production-Ready: Focused on real-world workflows, not just chat-based demos.
• Modular Architecture: Pluggable tools and agents using Semantic Kernel.
• Scalable: API-first design containerized with Docker and Redis.
• Secure by Design: Strict environment isolation and PII masking for memory.

---

📚 Technology Stack

• LLM: Azure OpenAI (GPT-4o)
• RAG: Azure AI Search
• Orchestration: Semantic Kernel v1.x
• API: FastAPI / Uvicorn
• Throttling: Redis
• Database: SQLite (Persistent via Docker Volumes)

---

🤝 Work with Bravado Solutions

We help enterprises move from AI experimentation to production-grade intelligent systems. Our team specializes in Agentic RAG, Cloud-Native SaaS, and Enterprise AI Orchestration.

• Portfolio Case Study: Detailed Agentic AI System - Empire Limousine

🌐 bravadosolutions.com
📧 contact@bravadosolutions.com

---

Built with ❤️ by Bravado Solutions.