Product
Explainable by design. Repeatable by principle.
Quality engineers are right to distrust black-box AI on a production line. J1 is built so you don't have to trust one.
The hybrid principle
AI finds the feature. A rule makes the call.
AI locates the feature. A deterministic rule makes the pass/fail decision. The result is explainable and repeatable — the same part gives the same answer, every time.
That split matters. A pure black-box model gives you a confidence score and no recourse when it's wrong. J1 uses AI only to locate features in the image; the pass/fail judgment is made by deterministic rules on geometric relationships that are independent of camera distance. When a part fails, the rule that failed it is on record — with the image. That's a decision you can defend in an audit, not a probability you have to apologize for.
Coverage
The visual defects that matter on a harness line.
- Connector presence and seating
- Locking tabs
- Retainers
- CPA seating
- Clip type and orientation
- Strap and band presence
Fail-closed by design
Nothing passes unless it is positively confirmed good.
Most inspection systems fail open: what they don't recognize, they wave through. J1 does the opposite. Any part the system cannot confirm — a defect it recognizes, or anything it can't classify — halts at the operator station and is flagged to a supervisor with photo evidence before the line continues. The unknown is treated as a defect until a person says otherwise. That's poka-yoke thinking applied to vision.
Traceability
Every inspection, logged with an image.
Each part inspected leaves a record: the image, the result, the time. When a customer concern comes in, you can pull the evidence for the exact part — not a sample, not a shift average. Built with IATF 16949 audit expectations in mind: objective evidence, retained and retrievable.
Deployment
A station at the operator's position. Nothing leaves the plant.
J1 runs as a self-contained station at the operator's position. Inspection runs on-site, at the edge — no cloud dependency, no images leaving your network for a part to pass. It fits into the station's cycle rather than adding a process step downstream.
Scope, honestly
What it doesn't do.
J1 catches visible and geometric defects — the escapes your dimensional checks pass. It does not claim to detect internal defects a camera cannot see, and we won't pretend otherwise. It is one layer of defense, aimed precisely at the layer your current controls miss.