Skip to content

README.md

Latest commit

 

History

History
161 lines (118 loc) · 8.53 KB

README.md

File metadata and controls

161 lines (118 loc) · 8.53 KB
Raw

Edge AI Demo Studio

Edge AI Demo Studio is a modern toolkit for deploying, managing, and serving AI models on edge platforms. It features a web-based UI for device, workload, and user management, and is optimized for Intel hardware and edge environments.

Demo

Features

  • Easy Model Deployment: Download, convert, and serve Hugging Face models with minimal setup.
  • Web-Based Management: Manage devices, workloads, and users through a modern web interface.
  • Edge Optimized: Built for Intel hardware and edge environments.
  • AI Services: AI Services that users can use for their applications
    • Text Generation — Generate coherent text using large language models optimized for Intel hardware.
    • Text to Speech — Natural-sounding speech synthesis with multiple voice options powered by Kokoro TTS.
    • Speech to Text — Real-time speech recognition with low-latency transcription using Whisper models optimized for Intel hardware.
    • Speaker Diarization — Identify and label speakers in audio recordings using pyannote.audio speaker diarization models.
    • Text Embedding — Generate dense vector embeddings for semantic search and RAG pipelines.
    • Reranker — Rescore and rerank documents by relevance for improved search and RAG pipelines.
    • Vector Database — FAISS-based vector storage for knowledge base management, semantic search, and RAG pipelines.
    • Lipsync — Real-time avatar lip-syncing with Wav2Lip, streamed over WebRTC.
    • Image Generation — Generate images from text prompts using diffusion models accelerated with OpenVINO.
    • MCP Manager — Manage Model Context Protocol servers and their tool integrations.
    • Wake Word Detection — Detect custom wake words from microphone input and send webhook notifications on detection events.
  • Samples: Sample use cases that implement the AI services (see Exporting Samples to package a subset for standalone deployment)
    • Digital Avatar — Interact with an AI-powered avatar that combines real-time video with intelligent conversation.
    • Digital Avatar Lite — A lightweight animated robot avatar that brings conversations to life with responsive movements and expressions.
    • RAG Chatbot — Upload documents and chat with an AI that retrieves relevant context to answer your questions.
    • Medical Scribe — Automatically transcribe and diarize doctor-patient conversations, then generate structured SOAP notes.
    • Webcam Capture with VLM — Demonstrate the integration of webcam capture and Visual Language Model (VLM) for enhanced interaction.
    • AI Exam Marking — AI-powered exam marking using OCR and LLM to automatically grade test papers from images.
    • PowerPoint Translator — Translate PowerPoint presentations while preserving formatting using AI.
    • Geti Image Classification — Classify images using a local Intel Geti deployment and send feedback for continuous model improvement.
    • Synthetic Image Generation — Generate and edit synthetic images from base images in real-time for dataset augmentation.
    • Robotics AI — A demo showcasing the capabilities of Robotics AI, including real-time object detection and manipulation.
  • Suites: Curated industry-specific AI solution packages built on Intel Edge AI Suites
    • Manufacturing AI Suite — A comprehensive toolkit for building, deploying, and scaling AI applications in industrial environments. Enables real-time integration with optimized hardware for production workflow automation, workplace safety, defect detection, and asset tracking.
      • Pallet Defect Detection — Real-time pallet condition monitoring on warehouse video streams using DL Streamer Pipeline Server, OpenVINO inference, and WebRTC streaming.
    • Metro AI Suite — Accelerates application development for edge AI video safety, security, and smart city use cases. Includes OpenVINO™ toolkit, Deep Learning Streamer, and Intel® oneAPI Toolkit for media analytics and AI performance optimization.
      • Image-Based Video Search — Near real-time image-based similarity search over live video streams using YOLOv11 object detection, ResNet-50 feature extraction via DL Streamer, and Milvus vector indexing.
    • Retail AI Suite — Accelerates development of edge AI applications for retail environments, enabling intelligent automation for use cases such as self-checkout, loss prevention, and store analytics with optimized Intel hardware and the OpenVINO™ toolkit.
      • Loss Prevention — Real-time self-checkout loss prevention using object detection and analytics to identify mis-scans and suspicious activity at the point of sale.

Architecture Diagram

Archictecture Diagram

Software Requirements

  • Operating System:
    • Ubuntu 24.04 LTS (other Linux distributions may work, but are not officially supported)
    • Windows 11 24H1 (other versions may work, but this is the validated version)
  • Other: See install_dependencies.sh for required system packages.

Quick Start

1. Install System Dependencies (Linux only)

sudo ./install_dependencies.sh

2. Set Up Python & Node.js Dependencies

For Linux:

./setup.sh

For Windows (PowerShell/Command Prompt):

./setup_win.bat

This will:

  • Set up a Python virtual environment
  • Install Python and Node.js dependencies
  • Start the frontend

3. Start the App

For Linux:

./start.sh

For Windows (PowerShell/Command Prompt):

./start_win.bat

Once started, access the web UI at http://localhost:8080.

Exporting Samples

The Export Samples feature lets you produce a slim, self-contained copy of Demo Studio that contains only the sample(s) you select, together with the services and workers they depend on. The exported directory includes its own setup.sh / setup_win.bat and start.sh / start_win.bat scripts so it can be set up and run independently.

Via the Launcher Script

Run from the repository root — Node.js is bootstrapped automatically from thirdparty/ if not already installed. The script lists all available samples, lets you pick one or more by number, then prompts for output directory, optional dependencies, and dry-run preference.

For Linux:

./export.sh

For Windows (PowerShell/Command Prompt):

.\export.bat

Via the Frontend GUI

  1. Open the web UI and navigate to the Samples page (http://localhost:8080/samples).
  2. Click Select to export (top-right area of the page) to enter selection mode.
  3. Check the sample card(s) you want to export.
  4. Click the Export selected button (download icon) that appears in the selection toolbar.
  5. Review the resolved export plan in the dialog — it lists the required services, any optional services (toggle Include optional services on/off), and the workers that will be bundled.
  6. Click Export to download a .zip archive of the self-contained project.

Docker

See docker/README.md for docker guidelines.

Project Structure

applications.ai.tools.edge-ai-demo-studio/
├── electron/         # Electron app
├── frontend/         # Next.js web frontend
├── workers/          # Python/AI backend services
├── models/           # Downloaded/converted models
├── scripts/          # Utility scripts
├── install_dependencies.sh  # System dependency installer
├── setup.sh / setup.ps1     # Project setup scripts
└── README.md

Deployment

The app is deployed using Electron to package the application. See docs/DEPLOYMENT.md for deployment guidelines.

FAQ

Q: Why is Electron Skipped by default

This is because Electron is being used to create a packaged release only. If you need a packaged release, please refer to docs/DEPLOYMENT.md