The Model Isn’t the Product: Building Streaming Evidence Layers for Industrial AI

Most AI demos assume clean data, stable networks, and users who live inside dashboards. Factory floors do not work that way.

This session shares how IntelFactor.ai is building an edge-first manufacturing intelligence system for real-time quality inspection, where local devices run computer vision inference, operators receive PASS / REVIEW / FAIL decisions on the line, and the cloud receives structured inspection records, evidence, review actions, and telemetry for analytics and AI-assisted root-cause analysis.

The core theme is simple: in industrial AI, the model is not the product. The product is the evidence layer that turns physical events into trustworthy, reviewable, and actionable data streams.

We will walk through the technical architecture behind this system, including NVIDIA Jetson edge inference, camera-based inspection, local persistence, evidence capture, HTTP-based inspection sync, Kafka-style event modeling, telemetry streams, and cloud workflows for review and analysis. The talk will also cover where streaming fits, where it does not, and why forcing every workflow into a topic can turn a practical system into a distributed systems museum exhibit.

Attendees will learn how to design production AI systems that connect edge inference, structured events, human-in-the-loop review, and cloud analytics. Key takeaways include how to model inspections as events, how to separate continuous telemetry from discrete inspection records, how to preserve traceability, and how to build systems that continue operating when networks, operators, and reality behave badly, as they always do.

This aligns well with Current’s suggested themes around AI in production, real-world applications, operational excellence, and edge-based streaming architectures.