Ticket Assistant — AI Support Agent for a Transport Booking App

---

📌 Overview

Ticket Assistant is an AI-powered conversational agent designed to help end-users navigate a transport ticketing application.
It assists with:

• 🔍 Finding transport routes
• 📄 Understanding UI elements (buttons, pages, navigation flows)
• 🧭 Guiding users through the booking process
• 🔒 Enforcing strict domain limitations (no off-topic answers)

This is achieved through:

• Local LLM (via LM Studio)
• Retrieval-Augmented Generation (RAG) with FAISS
• Tool-calling (LangGraph)
• Real-time Spring Boot backend integration
• FastAPI as the public interface

---

🎯 Features

✔ Local LLM inference

Runs through LM Studio using an OpenAI-compatible API.

✔ RAG Search (FAISS)

Indexes frontend_knowledge_base.md to provide accurate UI/UX guidance.

✔ Tool-calling via LangGraph

• rag_search → Retrieves UI information
• get_route_options → Queries Spring Boot /api/routes/options

✔ Strict domain rules

The assistant refuses:

• off-topic questions
• programming questions
• insults or unsafe content
• jailbreak attempts

✔ Session-based Conversation

Managed internally using a LangGraph state machine.

---

🏗 Project Structure

app/
├─ api/
│ ├─ routes.py # /chat endpoint
│ ├─ schemas.py # request/response models
├─ agent/
│ ├─ graph.py # LangGraph agent + tool calling
├─ tools/
│ ├─ rag_tools.py # RAG tool
│ └─ route_tools.py # Spring Boot route tool
├─ llm.py # LLM + embeddings initialization
├─ vectorstore.py # FAISS index builder/loader
├─ config.py # loads environment variables
└─ main.py # FastAPI setup
docs/
faiss_index/ # auto-generated FAISS index
frontend_knowledge_base.md
.env

---

⚙️ Installation & Setup

1️⃣ Install Python dependencies

pip install -r requirements.txt

2️⃣ Start LM Studio

1.Load a model (recommended: qwen3-vl-4b)

2. Start the OpenAI Compatible Server

3. Set listening port: 1234

📄 .env Configuration

Create a .env file in the project root:

LM_BASE_URL=
LM_API_KEY=
MODEL_NAME=

EMB_MODEL=
DOCS_DIR=./docs
INDEX_DIR=./faiss_index
MD_KNOWLEDGE_BASE=

SPRING_BASE_URL=
USER_AGENT=

🚀 Running the Application

Start FastAPI:

uvicorn app.main:app --reload --port 8001

🌐 API Endpoints POST /chat

Main endpoint used to interact with the AI agent.

Request:

{
  "message": "How do I search for transport routes?",
  "session_id": "demo123"
}

Response:

  {
    "reply": "To search for routes, navigate to the top bar and select 'Find Route'...",
    "session_id": "demo123"
  }

Graph Logic:

Frontend User
    │
    ▼
FastAPI (/chat)
    │
    ▼
LangGraph Agent ─────────────> Tool: rag_search (FAISS)
    │                                │
    │                                └── citește din frontend_knowledge_base.md
    │
    └───────────────> Tool: get_route_options (Spring Boot)
                                     │
                                     └── interoghează /api/routes/options