Manufacturing · Visual quality inspection
Banao builds computer-vision quality inspection that grades every part on the line in real time — cracks, glaze flaws, colour drift, surface defects — and sorts on the result.
The model runs edge-side on your existing line cameras: no cloud latency, no rip-and-replace. It is in production today on ceramic tile, and the pattern transfers to steel surface, auto paint, and packaging.
RAK Ceramics— defect model deployed edge-side with auto-grading into the conveyor sort.
A vision deployment is not just a model. It is the model, the integration, and the floor-team adoption — we own all three.
We label thousands of images of your real defects — cracks, glaze, colour, geometry — and train to your grading rules, not a generic dataset.
Inference runs on the line, on your existing cameras where possible. Decisions land in milliseconds, and the line never waits on a network round-trip.
The grade drives your conveyor and PLC sort logic, so good and reject parts separate automatically instead of waiting for a human call.
Defect trends by line, shift, and raw-material batch — so quality and plant heads see the pattern, not just the part.
We integrate with 1990s PLCs, SCADA, and analog sensors via retrofit. The model cares about the image signal, not the age of the machine.
The floor team can override and correct; those corrections feed back, so the model improves with each shift instead of drifting.
Metrics shown dotted (··) are being finalised in our case-study metrics pack — published only once verified.
RAK Ceramics
Inspection by eye was fatigued, inconsistent, and slow. Banao trained a defect model on thousands of crack, glaze, and colour images and deployed it edge-side with auto-grading into the conveyor sort.
CP Plus
Banao adds a computer-vision layer to existing CCTV and line cameras for quality, safety, and compliance — extending hardware already on the floor rather than replacing it.
Banao runs a ~300-person engineering company on its own AI products. InterviewGod screens our own hires; Vikaas runs our own demand generation. A model that has to survive our operation is already hardened before it reaches your line.
We are not describing production AI from the outside. We run it every working day — which is exactly the standard we hold a shop-floor inspection model to.
A camera is not always the answer. We will tell you before you spend on one:
We don't quote a vision line off a brochure. We look at your actual parts first.
We audit a sample of your real images, test feasibility on your hardest defect classes, and hand back a baseline accuracy estimate and ROI maths — yours to keep. If you proceed, the Sprint is credited against the build.
Label, train to your grading rules, and integrate with your cameras, PLCs, and sort hardware. Imaging and data pipeline are part of the deliverable.
Edge deployment with operator override and a plant dashboard, plus floor-team change management. Operator corrections feed the model every shift.
How many defect images do you need to start?
Enough to cover your real defect classes — often a few thousand labelled images. Where you have little history, the Discovery Sprint establishes whether augmentation or staged collection gets you to a usable baseline.
Will it work with our existing cameras?
Usually, yes. Banao deploys edge-side on existing line cameras and CCTV where the imaging is workable, and only specifies new cameras where lighting or resolution genuinely blocks accuracy. The week-one audit settles this.
What defect types can the model catch?
Surface and geometric defects the camera can see — cracks, glaze and coating flaws, colour drift, dimensional and edge defects, missing or misplaced features. We train to your grading rules, not a generic defect set.
Does it integrate with our sorting hardware and PLCs?
Yes. The grade drives your conveyor and PLC sort logic directly, including retrofit onto older PLCs and SCADA. Integration is part of the build deliverable, not a separate project.
How do operators stay in control?
Operators can override any call, and their corrections feed back into the model. Floor-team change management is a non-negotiable deliverable — adoption, not just accuracy, is what makes inspection AI stick.
Bring a sample of your real parts and your toughest defect class. In 45 minutes we'll tell you whether vision inspection is worth building — and what it would take.
Book a 45-min scoping call