The demo below runs on sample data.
ValveVision reads valve nameplates into a searchable asset register
ValveVision. Vision AI that reads nameplates, castings and tags from up to three photos per valve, maps the data to ASME, API and ISO terms and checks it for consistency.
- Client
- An industrial valve manufacturer
- Location
- UAE
- Year
- 2026
- Services
- AI Agents & Workflow Automation, AI Development, Data & Analytics
- Industry
- Quality, MES and inspection software for manufacturers
- Stack
- React, Tailwind CSS, Framer Motion, Recharts, Gemini, PostgreSQL
- Status
- Interactive demo
Overview
ValveVision is a vision AI system ShichifukuTekx built for an industrial valve manufacturer. It reads nameplates, cast body markings and ID tags from up to three photos of one valve, resolves worn text, maps the data to ASME, API and ISO terminology, checks material against pressure class and stores each asset with field-level confidence scores.
Where they started
- 01
Valve data was captured by hand from metal nameplates and cast markings. Many plates were worn, painted over or partly hidden, and the same valve often produced different records depending on who read it.
- 02
Fields that matter for safety, such as pressure class, body material and size, were recorded in free text with no consistent naming.
- 03
There was no single register of scanned assets. Engineers could not filter by material, rating or standard without rebuilding a spreadsheet.
- 04
Nobody checked whether the captured specification made engineering sense, for example whether a material grade fit the stated pressure class.
What we built
We built the capture flow around how valves are marked. The system accepts up to three perspectives of a single valve, typically the nameplate, the body casting and the ID tag, and cross-references them.
- Multi-perspective OCRGemini vision models read etched, embossed and tagged text and resolve characters that are worn or obscured in one view
- Nomenclature mappingraw readings are mapped to ASME, API and ISO terms, grouped into asset identity, material matrix and operating envelope
- Asset registrya searchable, filterable register of every scanned valve with field-level confidence scores
- Consistency checksrules test the specification, such as material grade against pressure class, and flag records that need an engineer's review
Confidence scores drive the workflow. High-confidence fields go straight into the register; low-confidence fields are highlighted so an engineer confirms them before the record is used.
The front end is React with Tailwind CSS, Framer Motion and Recharts. Records are stored in PostgreSQL, ready for export into ERP or maintenance systems.
This is the kind of applied vision work our AI agents and data analytics teams scope in a discovery sprint.
What changed
- 3 perspectives, per valve. Nameplate, body casting and ID tag cross-referenced
- ASME, API, ISO, standards mapping. Readings normalised to industry nomenclature
- Field-level, confidence scores. Low-confidence fields routed to an engineer
- Rule-checked, specifications. Material grade tested against pressure class