AI Nameplate Data Extraction Solution

Software-as-a-service (SaaS) artificial intelligence (AI) solution will extract equipment nameplate data from photographs and produce human-validated asset master data records in Loudoun Water's SAP Asset Loader format.
- Solution shall provide, at minimum, the following capabilities:
1. Image capture. Capture of equipment nameplate photographs by field staff using Loudoun Water‑managed mobile devices (iOS and/or Android) and tablets, plus bulk upload of existing photographs from a desktop browser. Describe any offline capture capability and how captured images synchronize when connectivity is restored.
2. AI extraction. Automated extraction of nameplate attributes from photographs, including at minimum: manufacturer, model number, serial number, asset type, voltage, amperage, horsepower/capacity, phase, RPM, frame, and year of manufacture. Offerors shall state the complete set of attributes their solution extracts and how additional Loudoun Water‑specific attributes are configured.
3. Normalization to Loudoun Water data standards. Extracted values shall be normalized to the data standards defined in Appendix D. Values that fall outside a controlled value list shall be flagged for review, not written to the output as free text. Offerors shall describe how these standards are configured and how they are maintained as Loudoun Water revises them.
4. Accuracy. The solution shall achieve a field‑level extraction accuracy of no less than 95% on legible nameplates. Fields that cannot be read shall be left empty and flagged for review, not estimated. Offerors shall state their measured accuracy, the method and dataset used to measure it, and how accuracy is monitored in production.
5. Confidence scoring and human‑in‑the‑loop review. Every extracted field shall carry a confidence indicator. Records or fields below a configurable confidence threshold shall be routed to a review queue for human verification and correction before release. No record shall reach Loudoun Water's Asset Loader output without passing the configured review workflow. The solution shall distinguish AI-generated values from human-verified values.
6. Edge case handling. Describe how the solution handles damaged, corroded, painted‑over, or partially illegible nameplates; poor lighting and glare; skewed or angled photographs; dirty equipment; multiple nameplates on a single asset; and a single nameplate photographed multiple times.
7. Duplicate detection. Automated detection of duplicate submissions using image characteristics and extracted values (for example serial number and manufacturer),with a defined resolution workflow
9. Image and document retention. Nameplate photographs shall be retained and associated with the resulting asset record. Offerors shall describe supported storage options (within the solution, export to Loudoun Water storage, or attachment to SAP Document Info Records), retention configuration, and export of the full image set on request.
10. Security and access control. Role‑based access control; authentication via Microsoft Entra ID (single sign‑on) with multi‑factor authentication; encryption of data in transit and at rest.
11. Audit logging. A complete, exportable audit trail covering image submission, extraction results, confidence scores, reviewer identity, every change made during review, and export events.
12.Capacity and throughput. Offerors shall state included annual extraction volume, concurrent user limits, batch size limits, typical processing time per image, and the pricing and mechanism for additional capacity.
13. Reporting. Dashboards and exportable reports covering extraction accuracy, data completeness, records created, review queue volume and aging, and per‑user field productivity, sufficient to support executive reporting.