Simplifying Gen AI Video Search & Summarization Setup with Docker Compose - Full Demo Breakdown

[vgygce19]

This discussion covers the key concepts demonstrated in the Simplifying AI Video Search & Summarization Setup with Docker Compose Video. The demo walks through the full end-to-end workflow of deploying an AI-powered Video summarization service on an Intel® Arrow Lake Core Ultra edge platform using the Edge AI Libraries and Edge Microvisor Toolkit (EMT) image.

🎥 Video Demo: https://youtu.be/yikAk5XU96s

The demo presents a complete edge-ready GenAI video summarization pipeline, including:

• A microservices-based architecture (video ingestion, summarization API, UI, object detection, text embedding, and model services)
• Deployment using Docker Compose for containerized orchestration
• Optimized inference using OpenVINO and Optimum Intel for VLM (Vision-Language Models), LLM, and object detection models
• A web interface for uploading videos, configuring chunk-based analysis, and viewing summaries

This application represents a turnkey pattern for running video-centric AI workloads at the edge and addresses common challenges in edge GenAI deployments:

✅ Challenge 1: Running VLMs and LLMs efficiently on edge hardware
➡️ Solution: INT8 quantization + OpenVINO + Optimum Intel accelerate inference while minimizing resource usage across CPU, GPU, and NPU.

✅ Challenge 2: Deploying complex multimodal GenAI pipelines
➡️ Solution: Microservices + Docker Compose simplify packaging, scaling, and operations for video summarization, object detection, and embeddings.

✅ Challenge 3: Handling large video files and enabling fast summarization
➡️ Solution: Chunk-based video processing with configurable frame intervals and ingestion settings ensures efficient analysis and summary generation.

✅ Challenge 4: Enabling non-technical users to access AI capabilities
➡️ Solution: Clean, simple, web-based UI for video uploads, pipeline configuration, and summary visualization.

✅ Challenge 5: Managing model execution reliably on edge nodes
➡️ Solution: EMT image + controlled container environment ensures repeatable deployments with environment variables for registry, credentials, and model configs.

Key Environment Variables Used

Registry & Credentials:
REGISTRY_URL, TAG, MINIO_ROOT_USER, MINIO_ROOT_PASSWORD, POSTGRES_USER, POSTGRES_PASSWORD, RABBITMQ_USER, RABBITMQ_PASSWORD

Models:

• VLM: Qwen/Qwen2.5-VL-3B-Instruct or microsoft/Phi-3.5-vision-instruct
• LLM: Intel/neural-chat-7b-v3-3
• Whisper ASR: tiny.en, small.en, medium.en
• Object Detection: yolov8l-worldv2
• Embeddings: CLIP/clip-vit-b-32, QwenText/qwen3-embedding-0.6b

Pipeline Config:
VS_WATCHER_DIR, FRAME_INTERVAL=15, OV_CONFIG='{"PERFORMANCE_HINT": "THROUGHPUT"}'

Key Workflow Steps

1. Preparing the Environment

• Verified platform details and Docker service status
• Cloned the Edge AI Libraries repository
• Navigated to the video-summarization sample folder
• Exported environment variables (registry details, credentials, model configs for summarization, object detection, and text embedding)
• Cleaned up any previous environment artifacts before starting

2. Installing Dependencies
The setup script installs:

• transformers and sentence-transformers for NLP tasks
• OpenVINO optimizations for hardware acceleration
• Optimum Intel runtime for efficient inference on Intel platforms
• Ensures optimized performance on Intel Arrow Lake Core Ultra with NPU, iGPU, and CPU orchestration

3. Pipeline Deployment

• Deploys a unified video search and summarization pipeline
• Optimizes YOLO object detection model for OpenVINO
• Launches all required microservices using Docker Compose
• Builds a production-ready containerized architecture

4. Video Processing and Summarization

• Upload videos via the web UI
• Configure pipeline settings:
• Define chunk duration
• Adjust ingestion parameters
• System processes selected chunks and generates concise summaries

5. Chunk-Based Analysis

• Splits video into segments for efficient processing
• Extracts key frames from each chunk
• Allows users to generate summaries for specific segments
• Improves speed, organization, and accuracy of video analysis

Main Takeaways

• Purpose-built for edge inference
• Optimized for Intel Arrow Lake Core Ultra with heterogeneous compute orchestration
• End-to-end containerized microservice architecture
• Full automation through setup scripts – no manual dependency handling
• Chunk-based summarization enables faster and more structured video analysis

Simple workflow: Upload → Configure → Summarize in seconds

👉 Learn more about the Simplifying AI Video Search & Summarization Setup with Docker Compose and Edge Microvisor Toolkit (EMT)
👉 Visit the Open Edge Platform Playlist for more demos

[biapalmeiro]

That is a great way of seeing different Open Edge Platform ingredients working together. Thanks for putting this together, @vgygce19