manufacturing-app-recipe
Installation
SKILL.md
Manufacturing App Recipe — vision / time-series / multimodal / VLM / agentic stacks on Intel hardware
Build a manufacturing quality/monitoring stack on Intel hardware with Docker Compose, in one of five deployment modes (Question 0 picks the shape):
| Mode | Shape | Components added on top of the previous mode |
|---|---|---|
vision |
lightweight single vision pipeline (demo/PoC only — no full stack) | DLSPS or a custom GStreamer pipeline |
ts |
sensor-only anomaly detection | Telegraf/InfluxDB + Time Series Analytics Microservice (Kapacitor UDF) + Grafana, as a new apps/<name>/ folder (or a bare UDF in demo mode) |
multimodal |
vision + sensor, one fused verdict | DLSPS + the ts stack + Fusion Analytics correlator, as a new flat stack directory |
vllm |
multimodal + narrative explanation |
an LLM/VLM (OVMS/vLLM) that explains each fused alert in natural language |
agentic |
vllm + full agent orchestration |
agent-quality-handler + model-download + (optional) metrics-manager microservices orchestrating the VLM layer, instead of a bare OVMS call |