Xana AI - Industrial Machine Support Assistant

Xana AI is an intelligent chatbot platform designed for shop-floor operators and technicians to interact with industrial machines. Built with Next.js (frontend) and NestJS (backend), it leverages RAG (Retrieval-Augmented Generation), vector embeddings, and LLM-powered conversational AI to provide contextual machine support, live data visualization, and alert monitoring.

---

🚀 Features

Frontend (Next.js)

• Conversational Chat Interface: Interactive chat UI with markdown support and syntax highlighting
• Multi-Asset Selection: Choose specific machines or query across all available assets
• Live Data Visualization: Real-time chart rendering using Chart.js for time-series metrics
• Alert Monitoring: Display machine alerts with severity, status, and timestamps
• Theme Support: Dark and light mode toggle for user preference
• Authentication: Token-based authentication integrated with IFF (IndustryFusion) suite
• Responsive Design: Built with Tailwind CSS and Radix UI components

Backend (NestJS)

• RAG-Powered Query Service: Semantic search using Milvus vector database with BGE-M3 embeddings
• LLM Integration: Meta LLaMA 3.3 70B Instruct model via IONOS Cloud API & Qwen2.5-14B-Instruct-fp16-ov via OpenVINO model server running on Intel dGPU like Battlemage or on CPU
• Intent Detection: Automatically detects chart and alert requests using structured LLM outputs
• Live Data Fetching: PostgreSQL TimescaleDB integration for historical machine metrics
• Alert Integration: Real-time alert retrieval from Alerta API
• Vector Store Management: MongoDB-based asset-to-vector-store mapping
• Security: JWT token handling with encryption/masking for sensitive data
• CORS & API Gateway: Configurable CORS and REST endpoints

---

📁 Project Structure

XanaAI/
├── backend/                    # NestJS REST API
│   ├── src/
│   │   ├── endpoints/
│   │   │   ├── query/         # Main query service with RAG
│   │   │   ├── ionos-rest/    # LLM & embedding API client IONOS
|   |   |   ├── opea-rest      # LLM & embedding API using OpenVINO server running on Intel 
|   |   |   ├── ollama-rest      # LLM & embedding API client using Ollama running on Intel
│   │   │   └── vector_mapping/ # Asset-to-vector store mapping
│   │   ├── data/jsonld/       # JSON-LD machine schemas
│   │   └── main.ts            # App entry (port 4050)
│   └── package.json
│
├── frontend/                   # Next.js application
│   ├── src/
│   │   ├── app/
│   │   │   └── page.tsx       # Main chat interface
│   │   ├── components/
│   │   │   ├── PromptBox.tsx  # User input component
│   │   │   └── AlertSummaryBlock.tsx
│   │   └── utility/tools.ts   # Helper functions
│   └── package.json
│
└── README.md

---

🛠️ Tech Stack

Backend

• Framework: NestJS (Node.js)
• LLM: Meta LLaMA 3.3 70B Instruct (via IONOS Cloud) OR Qwen2.5-14B-Instruct-fp16-ov via OpenVINO model server running on Intel dGPU like Battlemage or 0n CPU
• Embeddings: BAAI/bge-m3 (1024-dim vectors)
• Vector DB: Milvus (semantic search)
• Time-Series DB: PostgreSQL/TimescaleDB
• Alert System: Alerta API
• Metadata Store: MongoDB
• Authentication: JWT with JOSE encryption

Frontend

• Framework: Next.js 15 (React 18)
• Styling: Tailwind CSS 4, Radix UI
• Charts: Chart.js, PrimeReact
• Markdown: react-markdown with remark-gfm
• HTTP Client: Axios

---

⚙️ Setup & Installation

Prerequisites

• Node.js 20+
• PostgreSQL (TimescaleDB)
• MongoDB
• Milvus vector database
• Alerta instance (optional, for alerts)

Backend Setup

1. Navigate to backend directory

   cd backend

2. Install dependencies

   npm install

3. Configure environment variables looking .env.example
   Create a .env file:

   # API Keys
   COMPLETIONS_API_KEY=your_ionos_api_key
   COMPLETIONS_API_URL=https://inference.de-txl.ionos.com
   
   #OPEA OVMS Configuration (when LLM_PROVIDER="opea-ovms")
   OPEA_LLM_URL=http://localhost:8000/v3/chat/completions
   OPEA_LLM_MODEL=Qwen2.5-14B-Instruct-fp16-ov
   OPEA_CHAT_TIMEOUT=1800000  # 30 minutes
   
   # PostgreSQL (TimescaleDB)
   PGHOST=your_postgres_host
   PGPORT=5432
   PGPASSWORD=your_password
   PG_TABLE=entityhistory
   PGSSL=true
   
   # MongoDB
   MONGODB_URI=mongodb://localhost:27017
   MONGODB_DB=admin
   MONGODB_COL=vector_store_mappings
   
   # Milvus
   MILVUS_COLLECTION_NAME=custom_setup_6
   RAG_EMBED_DIM=1024
   
   # Alerta
   ALERTA_API_URL=https://alerta.example.com/api/alerts
   ALERTA_API_KEY=your_alerta_key
   
   # Security
   SECRET_KEY=your_jwt_secret
   MASK_SECRET=your_mask_secret
   REGISTRY_URL=https://registry.example.com
   
   # CORS
   CORS_ORIGIN=http://localhost:3050

4. Start development server

   npm run start:dev

   • Backend runs on http://localhost:4050

Frontend Setup

1. Navigate to frontend directory

   cd frontend

2. Install dependencies

   npm install

3. Configure environment variables
   Create a .env.local file:

   NEXT_PUBLIC_API_BASE=http://localhost:4050

4. Start development server

   npm run dev

   • Frontend runs on http://localhost:3050

---

🔧 Development Notes

Skip Authentication (Dev Mode)

In frontend/src/app/page.tsx (line ~70), change:

setLogin(false) → setLogin(true)

Key API Endpoints

Backend Routes:

• POST /query - Main chat query with RAG
• GET /vector-mappings - List available assets
• POST /auth/get-indexed-db-data - Retrieve indexed user data
• POST /ai/chat - Direct LLM completion (for testing)

Frontend Flow:

1. User authenticates via IFF token (URL param)
2. Loads available machines from /vector-mappings
3. Sends messages to /query with selected assets
4. Displays LLM response, charts, and alerts

---

🧠 How It Works

RAG Pipeline

1. User Query → Sent to backend with conversation history
2. Intent Detection → LLM determines if chart/alert data is needed
3. Vector Search → User question embedded → Milvus retrieves relevant docs
4. Context Injection → Search results added to system prompt
5. LLM Response → LLaMA generates answer using machine docs + context
6. Live Data → If chart/alert intent detected, fetches from Postgres/Alerta
7. Frontend Rendering → Displays text + charts + alerts

🚀 Production Deployment

Backend

npm run build
npm run start:prod

Frontend

npm run build
npm run start

Docker Support

Dockerfiles are included in both backend/ and frontend/ directories.

---

📝 License

This project is licensed under the terms specified in the LICENSE file.

---

🤝 Contributing

Developed and maintained by IndustryFusion.
For issues or feature requests, please contact the development team.

---

🔗 Related Resources

• NestJS Documentation
• Next.js Documentation
• Milvus Vector Database
• IONOS Cloud AI
• Meta LLaMA Models