How intel/edge-developer-kit-reference-scripts works
Project type and purpose
• This is a monorepo of _reference setup scripts_ and _sample use cases_ for various Intel platforms, GPUs, and workloads (AI, real‐time).
• It is not a single full‐stack app but a _toolkit_ that helps developers provision hardware/software (platform & GPU scripts) and deploy example workloads (use-cases).
Main components
a. Setup Modules (Shell scripts, README)
• Platform scripts (under /platforms) for Intel Atom, Core, Xeon, etc.
• GPU scripts (under /gpu) for Intel Arc and Flex series GPUs.
b. Use-Case Workloads (under /usecases)
• AI Use Cases
– OpenVINO examples, video-analytics app, Digital Avatar (full-stack), microservices (speech-to-text, TTS, text-to-image, vision-language, RAG toolkit)
– Digital Avatar splits into multiple Dockerized backends + Next.js frontend + Payload CMS
– Smart Parking and Vision Edge AI include cameras → inference → InfluxDB/Grafana dashboards
– LLM finetuning benchmarks, microservices
• Real-Time Use Cases
– TCC Tutorial: MQTT (Mosquitto) → Telegraf → InfluxDB → Grafana
Relationships and data flows
• Host/VM runs _setup scripts_ (platform + GPU) to install drivers, toolkits, OpenVINO.
• Once the host is provisioned, you select a use case. Most use cases ship a _docker-compose.yml_ to spin up:
• Container network (bridge)
• Inference microservices (each in its own container)
• Data stores (Chroma DB, Payload DB, InfluxDB)
• Messaging / broker (for real-time via MQTT)
• Web UI or API gateways (Next.js, FastAPI, Flask)
• Within Digital Avatar:
• Frontend (Next.js + Payload CMS) ←→ Backend API gateway ←→ {RealESRGan, LivePortrait, RAG-backend, SadTalker, Wav2Lip}
• Persistent volumes for model weights, assets
• Monitoring use cases (Smart Parking, Vision Edge AI, TCC tutorial) include
• Data collection modules → InfluxDB/Prometheus → Grafana dashboards
Architectural patterns and design principles
• Modular scripts per hardware/platform, each with its own setup.sh
• Containerized microservices—easy to swap or scale individual components
• Docker Compose orchestration—defines inter-service networking and dependencies
• Separation of concerns—platform provisioning, GPU provisioning, application deployment
Technologies, frameworks, and libraries
• Shell scripts (Bash) for environment provisioning
• Docker & Docker Compose
• Python (FastAPI, Flask, custom main.py in each service)
• Next.js & Payload CMS for full-stack UI (Digital Avatar, RAG toolkit)
• OpenVINO, Intel drivers—installed by platform/GPU scripts
• Databases: Chroma DB, Payload DB, InfluxDB, Prometheus
• Messaging: MQTT (Mosquitto)
Diagramming recommendations
• Blue for Setup components (scripts, driver install).
• Green for Container orchestration (Docker Compose).
• Purple for Data stores and Monitoring tools.
• Provisioning flow: scripts → host
• Network flow: frontend → API → inference containers → DB → frontend
• Monitoring flow: sensors → agent → DB → Grafana
Detail
• High-level module separately but allow “zoomed-in” callouts for complex use cases (e.g., Digital Avatar).
• Do not omit:
• Platform/GPU script directories and their mapping to hardware targets
• Use case orchestration (docker-compose, volumes, env files)
• Inter-container and container-to-host dependencies
• Monitoring and data-flow paths