Modern AI applications heavily depend on cloud infrastructure, external APIs, and expensive GPU resources. But what happens when AI needs to run inside a factory where internet connectivity is unreliable, latency matters, and sensitive production data cannot leave the premises?
In this talk, we'll explore how to build an offline-first Industrial AI architecture that brings intelligence directly to the factory floor using open-source technologies.
We'll walk through a practical manufacturing scenario where an edge device captures product images, detects defects using computer vision, publishes machine events through MQTT, and uses locally running AI models to investigate production anomalies and generate actionable insights.
The session will cover:
Designing an offline-first Industrial IoT and Edge AI architecture.
Building computer vision pipelines for manufacturing defect detection.
Connecting industrial devices using MQTT and open protocols.
Running local AI models using Ollama without external AI APIs.
Integrating AI agents with machine telemetry through MCP.
Processing industrial events and generating actionable insights.
Designing resilient systems that continue operating without internet connectivity.
Rather than focusing only on theoretical architecture, this session will demonstrate how different open-source technologies can work together to build a practical Industrial AI solution.
Attendees will leave with an understanding of how to move beyond cloud-dependent AI applications and bring intelligent capabilities closer to real-world industrial systems.